SUMMARY: Over 55% of author names in PubMed are ambiguous: the same name is shared by different individual researchers. This poses significant challenges on precise literature retrieval for author name queries, a common behavior in biomedical literature search. In response, we present a comprehensive dataset of disambiguated authors. Specifically, we complement the automatic PubMed Computed Authors algorithm with the latest ORCID data for improved accuracy. As a result, the enhanced algorithm achieves high performance in author name disambiguation, and subsequently our dataset contains more than 21 million disambiguated authors for over 35 million PubMed articles and is incrementally updated on a weekly basis. More importantly, we make the dataset publicly available for the community such that it can be utilized in a wide variety of potential applications beyond assisting PubMed's author name queries. Finally, we propose a set of guidelines for best practices of authors pertaining to use of their names. AVAILABILITY AND IMPLEMENTATION: The PubMed Computed Authors dataset is publicly available for bulk download at: https://ftp.ncbi.nlm.nih.gov/pub/lu/ComputedAuthors/. Additionally, it is available for query through web API at: https://www.ncbi.nlm.nih.gov/research/bionlp/APIs/authors/.

MOTIVATION: Information retrieval (IR) is essential in biomedical knowledge acquisition and clinical decision support. While recent progress has shown that language model encoders perform better semantic retrieval, training such models requires abundant query-article annotations that are difficult to obtain in biomedicine. As a result, most biomedical IR systems only conduct lexical matching. In response, we introduce MedCPT, a first-of-its-kind Contrastively Pre-trained Transformer model for zero-shot semantic IR in biomedicine. RESULTS: To train MedCPT, we collected an unprecedented scale of 255 million user click logs from PubMed. With such data, we use contrastive learning to train a pair of closely integrated retriever and re-ranker. Experimental results show that MedCPT sets new state-of-the-art performance on six biomedical IR tasks, outperforming various baselines including much larger models, such as GPT-3-sized cpt-text-XL. In addition, MedCPT also generates better biomedical article and sentence representations for semantic evaluations. As such, MedCPT can be readily applied to various real-world biomedical IR tasks. AVAILABILITY AND IMPLEMENTATION: The MedCPT code and model are available at https://github.com/ncbi/MedCPT.

OBJECTIVE: A significant number of recent articles in PubMed have full text available in PubMed Central®, and the availability of full texts has been consistently growing. However, it is not currently possible for a user to simultaneously query the contents of both databases and receive a single integrated search result. In this study, we investigate how to score full text articles given a multitoken query and how to combine those full text article scores with scores originating from abstracts and achieve an overall improved retrieval performance. MATERIALS AND METHODS: For scoring full text articles, we propose a method to combine information coming from different sections by converting the traditionally used BM25 scores into log odds ratio scores which can be treated uniformly. We further propose a method that successfully combines scores from two heterogenous retrieval sources - full text articles and abstract only articles - by balancing the contributions of their respective scores through a probabilistic transformation. We use PubMed click data that consists of queries sampled from PubMed user logs along with a subset of retrieved and clicked documents to train the probabilistic functions and to evaluate retrieval effectiveness. RESULTS AND CONCLUSIONS: Random ranking achieves 0.579 MAP score on our PubMed click data. BM25 ranking on PubMed abstracts improves the MAP by 10.6%. For full text documents, experiments confirm that BM25 section scores are of different value depending on the section type and are not directly comparable. Naïvely using the body text of articles along with abstract text degrades the overall quality of the search. The proposed log odds ratio scores normalize and combine the contributions of occurrences of query tokens in different sections. By including full text where available, we gain another 0.67%, or 7% relative improvement over abstract alone. We find an advantage in the more accurate estimate of the value of BM25 scores depending on the section from which they were produced. Taking the sum of top three section scores performs the best.

OBJECTIVE: In a biomedical literature search, the link between a query and a document is often not established, because they use different terms to refer to the same concept. Distributional word embeddings are frequently used for detecting related words by computing the cosine similarity between them. However, previous research has not established either the best embedding methods for detecting synonyms among related word pairs or how effective such methods may be. MATERIALS AND METHODS: In this study, we first create the BioSearchSyn set, a manually annotated set of synonyms, to assess and compare 3 widely used word-embedding methods (word2vec, fastText, and GloVe) in their ability to detect synonyms among related pairs of words. We demonstrate the shortcomings of the cosine similarity score between word embeddings for this task: the same scores have very different meanings for the different methods. To address the problem, we propose utilizing pool adjacent violators (PAV), an isotonic regression algorithm, to transform a cosine similarity into a probability of 2 words being synonyms. RESULTS: Experimental results using the BioSearchSyn set as a gold standard reveal which embedding methods have the best performance in identifying synonym pairs. The BioSearchSyn set also allows converting cosine similarity scores into probabilities, which provides a uniform interpretation of the synonymy score over different methods. CONCLUSIONS: We introduced the BioSearchSyn corpus of 1000 term pairs, which allowed us to identify the best embedding method for detecting synonymy for biomedical search. Using the proposed method, we created PubTermVariants2.0: a large, automatically extracted set of synonym pairs that have augmented PubMed searches since the spring of 2019.

The need to organize any large document collection in a manner that facilitates human comprehension has become crucial with the increasing volume of information available. Two common approaches to provide a broad overview of the information space are document clustering and topic modeling. Clustering aims to group documents or terms into meaningful clusters. Topic modeling, on the other hand, focuses on finding coherent keywords for describing topics appearing in a set of documents. In addition, there have been efforts for clustering documents and finding keywords simultaneously. We present an algorithm to analyze document collections that is based on a notion of a theme, defined as a dual representation based on a set of documents and key terms. In this work, a novel vector space mechanism is proposed for computing themes. Starting with a single document, the theme algorithm treats terms and documents as explicit components, and iteratively uses each representation to refine the other until the theme is detected. The method heavily relies on an optimization routine that we refer to as the projection algorithm which, under specific conditions, is guaranteed to converge to the first singular vector of a data matrix. We apply our algorithm to a collection of about sixty thousand PubMed Ⓡ documents examining the subject of Single Nucleotide Polymorphism, evaluate the results and show the effectiveness and scalability of the proposed method. This study presents a contribution on theoretical and algorithmic levels, as well as demonstrates the feasibility of the method for large scale applications. The evaluation of our system on benchmark datasets demonstrates that our method compares favorably with the current state-of-the-art methods in computing clusters of documents with coherent topic terms.

UNLABELLED: Medical Subject Headings (MeSH(®)) is a controlled vocabulary for indexing and searching biomedical literature. MeSH terms and subheadings are organized in a hierarchical structure and are used to indicate the topics of an article. Biologists can use either MeSH terms as queries or the MeSH interface provided in PubMed(®) for searching PubMed abstracts. However, these are rarely used, and there is no convenient way to link standardized MeSH terms to user queries. Here, we introduce a web interface which allows users to enter queries to find MeSH terms closely related to the queries. Our method relies on co-occurrence of text words and MeSH terms to find keywords that are related to each MeSH term. A query is then matched with the keywords for MeSH terms, and candidate MeSH terms are ranked based on their relatedness to the query. The experimental results show that our method achieves the best performance among several term extraction approaches in terms of topic coherence. Moreover, the interface can be effectively used to find full names of abbreviations and to disambiguate user queries. AVAILABILITY AND IMPLEMENTATION: https://www.ncbi.nlm.nih.gov/IRET/MESHABLE/ CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Improving the search and browsing experience in PubMed is a key component in helping users detect information of interest.In particular, when exploring a novel field, it is important to provide a comprehensive view for a specific subject.One solution for providing this panoramic picture is to find sub-topics from a set of documents.We propose a method that finds sub-topics that we refer to as themes and computes representative titles based on a set of documents in each theme.The method combines a thematic clustering algorithm and the Pool Adjacent Violators algorithm to induce significant themes.Then, for each theme, a title is computed using PubMed document titles and theme-dependent term scores.We tested our system on five disease sets from OMIM and evaluated the results based on normalized point-wise mutual information and MeSH terms.For both performance measures, the proposed approach outperformed LDA.The quality of theme titles were also evaluated by comparing them with manually created titles.

BACKGROUND: Controlled vocabularies such as the Unified Medical Language System (UMLS) and Medical Subject Headings (MeSH) are widely used for biomedical natural language processing (NLP) tasks. However, the standard terminology in such collections suffers from low usage in biomedical literature, e.g. only 13% of UMLS terms appear in MEDLINE. RESULTS: We here propose an efficient and effective method for extracting noun phrases for biomedical semantic categories. The proposed approach utilizes simple linguistic patterns to select candidate noun phrases based on headwords, and a machine learning classifier is used to filter out noisy phrases. For experiments, three NLP rules were tested and manually evaluated by three annotators. Our approaches showed over 93% precision on average for the headwords, "gene", "protein", "disease", "cell" and "cells". CONCLUSIONS: Although biomedical terms in knowledge-rich resources may define semantic categories, variations of the controlled terms in literature are still difficult to identify. The method proposed here is an effort to narrow the gap between controlled vocabularies and the entities used in text. Our extraction method cannot completely eliminate manual evaluation, however a simple and automated solution with high precision performance provides a convenient way for enriching semantic categories by incorporating terms obtained from the literature.

IN BIONLP-ST 2013: We participated in the BioNLP 2013 shared tasks on event extraction. Our extraction method is based on the search for an approximate subgraph isomorphism between key context dependencies of events and graphs of input sentences. Our system was able to address both the GENIA (GE) task focusing on 13 molecular biology related event types and the Cancer Genetics (CG) task targeting a challenging group of 40 cancer biology related event types with varying arguments concerning 18 kinds of biological entities. In addition to adapting our system to the two tasks, we also attempted to integrate semantics into the graph matching scheme using a distributional similarity model for more events, and evaluated the event extraction impact of using paths of all possible lengths as key context dependencies beyond using only the shortest paths in our system. We achieved a 46.38% F-score in the CG task (ranking 3rd) and a 48.93% F-score in the GE task (ranking 4th). AFTER BIONLP-ST 2013: We explored three ways to further extend our event extraction system in our previously published work: (1) We allow non-essential nodes to be skipped, and incorporated a node skipping penalty into the subgraph distance function of our approximate subgraph matching algorithm. (2) Instead of assigning a unified subgraph distance threshold to all patterns of an event type, we learned a customized threshold for each pattern. (3) We implemented the well-known Empirical Risk Minimization (ERM) principle to optimize the event pattern set by balancing prediction errors on training data against regularization. When evaluated on the official GE task test data, these extensions help to improve the extraction precision from 62% to 65%. However, the overall F-score stays equivalent to the previous performance due to a 1% drop in recall.

Identifying unknown drug interactions is of great benefit in the early detection of adverse drug reactions. Despite existence of several resources for drug-drug interaction (DDI) information, the wealth of such information is buried in a body of unstructured medical text which is growing exponentially. This calls for developing text mining techniques for identifying DDIs. The state-of-the-art DDI extraction methods use Support Vector Machines (SVMs) with non-linear composite kernels to explore diverse contexts in literature. While computationally less expensive, linear kernel-based systems have not achieved a comparable performance in DDI extraction tasks. In this work, we propose an efficient and scalable system using a linear kernel to identify DDI information. The proposed approach consists of two steps: identifying DDIs and assigning one of four different DDI types to the predicted drug pairs. We demonstrate that when equipped with a rich set of lexical and syntactic features, a linear SVM classifier is able to achieve a competitive performance in detecting DDIs. In addition, the one-against-one strategy proves vital for addressing an imbalance issue in DDI type classification. Applied to the DDIExtraction 2013 corpus, our system achieves an F1 score of 0.670, as compared to 0.651 and 0.609 reported by the top two participating teams in the DDIExtraction 2013 challenge, both based on non-linear kernel methods.

MOTIVATION: Clustering methods can be useful for automatically grouping documents into meaningful clusters, improving human comprehension of a document collection. Although there are clustering algorithms that can achieve the goal for relatively large document collections, they do not always work well for small and homogenous datasets. METHODS: In this article, we present Retro-a novel clustering algorithm that extracts meaningful clusters along with concise and descriptive titles from small and homogenous document collections. Unlike common clustering approaches, our algorithm predicts cluster titles before clustering. It relies on the hypergeometric distribution model to discover key phrases, and generates candidate clusters by assigning documents to these phrases. Further, the statistical significance of candidate clusters is tested using supervised learning methods, and a multiple testing correction technique is used to control the overall quality of clustering. RESULTS: We test our system on five disease datasets from OMIM(®) and evaluate the results based on MeSH(®) term assignments. We further compare our method with several baseline and state-of-the-art methods, including K-means, expectation maximization, latent Dirichlet allocation-based clustering, Lingo, OPTIMSRC and adapted GK-means. The experimental results on the 20-Newsgroup and ODP-239 collections demonstrate that our method is successful at extracting significant clusters and is superior to existing methods in terms of quality of clusters. Finally, we apply our system to a collection of 6248 topical sets from the HomoloGene(®) database, a resource in PubMed(®). Empirical evaluation confirms the method is useful for small homogenous datasets in producing meaningful clusters with descriptive titles. AVAILABILITY AND IMPLEMENTATION: A web-based demonstration of the algorithm applied to a collection of sets from the HomoloGene database is available at http://www.ncbi.nlm.nih.gov/CBBresearch/Wilbur/IRET/CLUSTERING_HOMOLOGENE/index.html. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Log analysis shows that PubMed users frequently use author names in queries for retrieving scientific literature. However, author name ambiguity may lead to irrelevant retrieval results. To improve the PubMed user experience with author name queries, we designed an author name disambiguation system consisting of similarity estimation and agglomerative clustering. A machine-learning method was employed to score the features for disambiguating a pair of papers with ambiguous names. These features enable the computation of pairwise similarity scores to estimate the probability of a pair of papers belonging to the same author, which drives an agglomerative clustering algorithm regulated by 2 factors: name compatibility and probability level. With transitivity violation correction, high precision author clustering is achieved by focusing on minimizing false-positive pairing. Disambiguation performance is evaluated with manual verification of random samples of pairs from clustering results. When compared with a state-of-the-art system, our evaluation shows that among all the pairs the lumping error rate drops from 10.1% to 2.2% for our system, while the splitting error rises from 1.8% to 7.7%. This results in an overall error rate of 9.9%, compared with 11.9% for the state-of-the-art method. Other evaluations based on gold standard data also show the increase in accuracy of our clustering. We attribute the performance improvement to the machine-learning method driven by a large-scale training set and the clustering algorithm regulated by a name compatibility scheme preferring precision. With integration of the author name disambiguation system into the PubMed search engine, the overall click-through-rate of PubMed users on author name query results improved from 34.9% to 36.9%.

MOTIVATION: Finding protein-protein interaction (PPI) information from literature is challenging but an important issue. However, keyword search in PubMed(®) is often time consuming because it requires a series of actions that refine keywords and browse search results until it reaches a goal. Due to the rapid growth of biomedical literature, it has become more difficult for biologists and curators to locate PPI information quickly. Therefore, a tool for prioritizing PPI informative articles can be a useful assistant for finding this PPI-relevant information. RESULTS: PIE (Protein Interaction information Extraction) the search is a web service implementing a competition-winning approach utilizing word and syntactic analyses by machine learning techniques. For easy user access, PIE the search provides a PubMed-like search environment, but the output is the list of articles prioritized by PPI confidence scores. By obtaining PPI-related articles at high rank, researchers can more easily find the up-to-date PPI information, which cannot be found in manually curated PPI databases. AVAILABILITY: http://www.ncbi.nlm.nih.gov/IRET/PIE/.

MOTIVATION: Research in the biomedical domain can have a major impact through open sharing of the data produced. For this reason, it is important to be able to identify instances of data production and deposition for potential re-use. Herein, we report on the automatic identification of data deposition statements in research articles. RESULTS: We apply machine learning algorithms to sentences extracted from full-text articles in PubMed Central in order to automatically determine whether a given article contains a data deposition statement, and retrieve the specific statements. With an Support Vector Machine classifier using conditional random field determined deposition features, articles containing deposition statements are correctly identified with 81% F-measure. An error analysis shows that almost half of the articles classified as containing a deposition statement by our method but not by the gold standard do indeed contain a deposition statement. In addition, our system was used to process articles in PubMed Central, predicting that a total of 52 932 articles report data deposition, many of which are not currently included in the Secondary Source Identifier [si] field for MEDLINE citations. AVAILABILITY: All annotated datasets described in this study are freely available from the NLM/NCBI website at http://www.ncbi.nlm.nih.gov/CBBresearch/Fellows/Neveol/DepositionDataSets.zip CONTACT: [email protected]; [email protected]; [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: The overall goal of the BioCreative Workshops is to promote the development of text mining and text processing tools which are useful to the communities of researchers and database curators in the biological sciences. To this end BioCreative I was held in 2004, BioCreative II in 2007, and BioCreative II.5 in 2009. Each of these workshops involved humanly annotated test data for several basic tasks in text mining applied to the biomedical literature. Participants in the workshops were invited to compete in the tasks by constructing software systems to perform the tasks automatically and were given scores based on their performance. The results of these workshops have benefited the community in several ways. They have 1) provided evidence for the most effective methods currently available to solve specific problems; 2) revealed the current state of the art for performance on those problems; 3) and provided gold standard data and results on that data by which future advances can be gauged. This special issue contains overview papers for the three tasks of BioCreative III. RESULTS: The BioCreative III Workshop was held in September of 2010 and continued the tradition of a challenge evaluation on several tasks judged basic to effective text mining in biology, including a gene normalization (GN) task and two protein-protein interaction (PPI) tasks. In total the Workshop involved the work of twenty-three teams. Thirteen teams participated in the GN task which required the assignment of EntrezGene IDs to all named genes in full text papers without any species information being provided to a system. Ten teams participated in the PPI article classification task (ACT) requiring a system to classify and rank a PubMed® record as belonging to an article either having or not having "PPI relevant" information. Eight teams participated in the PPI interaction method task (IMT) where systems were given full text documents and were required to extract the experimental methods used to establish PPIs and a text segment supporting each such method. Gold standard data was compiled for each of these tasks and participants competed in developing systems to perform the tasks automatically.BioCreative III also introduced a new interactive task (IAT), run as a demonstration task. The goal was to develop an interactive system to facilitate a user's annotation of the unique database identifiers for all the genes appearing in an article. This task included ranking genes by importance (based preferably on the amount of described experimental information regarding genes). There was also an optional task to assist the user in finding the most relevant articles about a given gene. For BioCreative III, a user advisory group (UAG) was assembled and played an important role 1) in producing some of the gold standard annotations for the GN task, 2) in critiquing IAT systems, and 3) in providing guidance for a future more rigorous evaluation of IAT systems. Six teams participated in the IAT demonstration task and received feedback on their systems from the UAG group. Besides innovations in the GN and PPI tasks making them more realistic and practical and the introduction of the IAT task, discussions were begun on community data standards to promote interoperability and on user requirements and evaluation metrics to address utility and usability of systems. CONCLUSIONS: In this paper we give a brief history of the BioCreative Workshops and how they relate to other text mining competitions in biology. This is followed by a synopsis of the three tasks GN, PPI, and IAT in BioCreative III with figures for best participant performance on the GN and PPI tasks. These results are discussed and compared with results from previous BioCreative Workshops and we conclude that the best performing systems for GN, PPI-ACT and PPI-IMT in realistic settings are not sufficient for fully automatic use. This provides evidence for the importance of interactive systems and we present our vision of how best to construct an interactive system for a GN or PPI like task in the remainder of the paper.

BACKGROUND: Identifying protein-protein interactions (PPIs) from literature is an important step in mining the function of individual proteins as well as their biological network. Since it is known that PPIs have distinctive patterns in text, machine learning approaches have been successfully applied to mine these patterns. However, the complex nature of PPI description makes the extraction process difficult. RESULTS: Our approach utilizes both word and syntactic features to effectively capture PPI patterns from biomedical literature. The proposed method automatically identifies gene names by a Priority Model, then extracts grammar relations using a dependency parser. A large margin classifier with Huber loss function learns from the extracted features, and unknown articles are predicted using this data-driven model. For the BioCreative III ACT evaluation, our official runs were ranked in top positions by obtaining maximum 89.15% accuracy, 61.42% F1 score, 0.55306 MCC score, and 67.98% AUC iP/R score. CONCLUSIONS: Even though problems still remain, utilizing syntactic information for article-level filtering helps improve PPI ranking performance. The proposed system is a revision of previously developed algorithms in our group for the ACT evaluation. Our approach is valuable in showing how to use grammatical relations for PPI article filtering, in particular, with a limited training corpus. While current performance is far from satisfactory as an annotation tool, it is already useful for a PPI article search engine since users are mainly focused on highly-ranked results.

BACKGROUND: Determining usefulness of biomedical text mining systems requires realistic task definition and data selection criteria without artificial constraints, measuring performance aspects that go beyond traditional metrics. The BioCreative III Protein-Protein Interaction (PPI) tasks were motivated by such considerations, trying to address aspects including how the end user would oversee the generated output, for instance by providing ranked results, textual evidence for human interpretation or measuring time savings by using automated systems. Detecting articles describing complex biological events like PPIs was addressed in the Article Classification Task (ACT), where participants were asked to implement tools for detecting PPI-describing abstracts. Therefore the BCIII-ACT corpus was provided, which includes a training, development and test set of over 12,000 PPI relevant and non-relevant PubMed abstracts labeled manually by domain experts and recording also the human classification times. The Interaction Method Task (IMT) went beyond abstracts and required mining for associations between more than 3,500 full text articles and interaction detection method ontology concepts that had been applied to detect the PPIs reported in them. RESULTS: A total of 11 teams participated in at least one of the two PPI tasks (10 in ACT and 8 in the IMT) and a total of 62 persons were involved either as participants or in preparing data sets/evaluating these tasks. Per task, each team was allowed to submit five runs offline and another five online via the BioCreative Meta-Server. From the 52 runs submitted for the ACT, the highest Matthew's Correlation Coefficient (MCC) score measured was 0.55 at an accuracy of 89% and the best AUC iP/R was 68%. Most ACT teams explored machine learning methods, some of them also used lexical resources like MeSH terms, PSI-MI concepts or particular lists of verbs and nouns, some integrated NER approaches. For the IMT, a total of 42 runs were evaluated by comparing systems against manually generated annotations done by curators from the BioGRID and MINT databases. The highest AUC iP/R achieved by any run was 53%, the best MCC score 0.55. In case of competitive systems with an acceptable recall (above 35%) the macro-averaged precision ranged between 50% and 80%, with a maximum F-Score of 55%. CONCLUSIONS: The results of the ACT task of BioCreative III indicate that classification of large unbalanced article collections reflecting the real class imbalance is still challenging. Nevertheless, text-mining tools that report ranked lists of relevant articles for manual selection can potentially reduce the time needed to identify half of the relevant articles to less than 1/4 of the time when compared to unranked results. Detecting associations between full text articles and interaction detection method PSI-MI terms (IMT) is more difficult than might be anticipated. This is due to the variability of method term mentions, errors resulting from pre-processing of articles provided as PDF files, and the heterogeneity and different granularity of method term concepts encountered in the ontology. However, combining the sophisticated techniques developed by the participants with supporting evidence strings derived from the articles for human interpretation could result in practical modules for biological annotation workflows.

BACKGROUND: We report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers of the genes detected in full-text articles. For training, 32 fully and 500 partially annotated articles were prepared. A total of 507 articles were selected as the test set. Due to the high annotation cost, it was not feasible to obtain gold-standard human annotations for all test articles. Instead, we developed an Expectation Maximization (EM) algorithm approach for choosing a small number of test articles for manual annotation that were most capable of differentiating team performance. Moreover, the same algorithm was subsequently used for inferring ground truth based solely on team submissions. We report team performance on both gold standard and inferred ground truth using a newly proposed metric called Threshold Average Precision (TAP-k). RESULTS: We received a total of 37 runs from 14 different teams for the task. When evaluated using the gold-standard annotations of the 50 articles, the highest TAP-k scores were 0.3297 (k=5), 0.3538 (k=10), and 0.3535 (k=20), respectively. Higher TAP-k scores of 0.4916 (k=5, 10, 20) were observed when evaluated using the inferred ground truth over the full test set. When combining team results using machine learning, the best composite system achieved TAP-k scores of 0.3707 (k=5), 0.4311 (k=10), and 0.4477 (k=20) on the gold standard, representing improvements of 12.4%, 21.8%, and 26.6% over the best team results, respectively. CONCLUSIONS: By using full text and being species non-specific, the GN task in BioCreative III has moved closer to a real literature curation task than similar tasks in the past and presents additional challenges for the text mining community, as revealed in the overall team results. By evaluating teams using the gold standard, we show that the EM algorithm allows team submissions to be differentiated while keeping the manual annotation effort feasible. Using the inferred ground truth we show measures of comparative performance between teams. Finally, by comparing team rankings on gold standard vs. inferred ground truth, we further demonstrate that the inferred ground truth is as effective as the gold standard for detecting good team performance.

Given prior human judgments of the condition of an object it is possible to use these judgments to make a maximal likelihood estimate of what future human judgments of the condition of that object will be. However, if one has a reasonably large collection of similar objects and the prior human judgments of a number of judges regarding the condition of each object in the collection, then it is possible to make predictions of future human judgments for the whole collection that are superior to the simple maximal likelihood estimate for each object in isolation. This is possible because the multiple judgments over the collection allow an analysis to determine the relative value of a judge as compared with the other judges in the group and this value can be used to augment or diminish a particular judge's influence in predicting future judgments. Here we study and compare five different methods for making such improved predictions and show that each is superior to simple maximal likelihood estimates.

BACKGROUND: The rapid growth of biomedical literature requires accurate text analysis and text processing tools. Detecting abbreviations and identifying their definitions is an important component of such tools. Most existing approaches for the abbreviation definition identification task employ rule-based methods. While achieving high precision, rule-based methods are limited to the rules defined and fail to capture many uncommon definition patterns. Supervised learning techniques, which offer more flexibility in detecting abbreviation definitions, have also been applied to the problem. However, they require manually labeled training data. METHODS: In this work, we develop a machine learning algorithm for abbreviation definition identification in text which makes use of what we term naturally labeled data. Positive training examples are naturally occurring potential abbreviation-definition pairs in text. Negative training examples are generated by randomly mixing potential abbreviations with unrelated potential definitions. The machine learner is trained to distinguish between these two sets of examples. Then, the learned feature weights are used to identify the abbreviation full form. This approach does not require manually labeled training data. RESULTS: We evaluate the performance of our algorithm on the Ab3P, BIOADI and Medstract corpora. Our system demonstrated results that compare favourably to the existing Ab3P and BIOADI systems. We achieve an F-measure of 91.36% on Ab3P corpus, and an F-measure of 87.13% on BIOADI corpus which are superior to the results reported by Ab3P and BIOADI systems. Moreover, we outperform these systems in terms of recall, which is one of our goals.

Given prior human judgments of the condition of an object it is possible to use these judgments to make a maximal likelihood estimate of what future human judgments of the condition of that object will be. However, if one has a reasonably large collection of similar objects and the prior human judgments of a number of judges regarding the condition of each object in the collection, then it is possible to make predictions of future human judgments for the whole collection that are superior to the simple maximal likelihood estimate for each object in isolation. This is possible because the multiple judgments over the collection allow an analysis to determine the relative value of a judge as compared with the other judges in the group and this value can be used to augment or diminish a particular judge's influence in predicting future judgments. Here we study and compare five different methods for making such improved predictions and show that each is superior to simple maximal likelihood estimates.

The rapid growth of biomedical literature requires accurate text analysis and text processing tools. Detecting abbreviations and identifying their definitions is an important component of such tools. In this work, we develop a machine learning algorithm for abbreviation definition identification in text. Most existing approaches for abbreviation definition identification employ rule-based methods. While achieving high precision, rule-based methods are limited to the rules defined and fail to capture many uncommon definition patterns. Supervised learning techniques, which offer more flexibility in detecting abbreviation definitions, have also been applied to the problem. However, they require manually labeled training data. In this study, we make use of what we term naturally labeled data. Positive training examples are extracted from text, which provides naturally occurring potential abbreviation-definition pairs. Negative training examples are generated randomly by mixing potential abbreviations with unrelated potential definitions. The machine learner is trained to distinguish between these two sets of examples. Then, the learned feature weights are used to identify the abbreviation full form. This approach does not require manually labeled training data. We evaluate the performance of our algorithm on the Ab3P, BIOADI and Meds tract corpora. We achieve an F-score that is comparable to the earlier existing systems yet with a higher recall.

We explore the feasibility of automatically identifying sentences in different MEDLINE abstracts that are related in meaning. We compared traditional vector space models with machine learning methods for detecting relatedness, and found that machine learning was superior. The Huber method, a variant of Support Vector Machines which minimizes the modified Huber loss function, achieves 73% precision when the score cutoff is set high enough to identify about one related sentence per abstract on average. We illustrate how an abstract viewed in PubMed might be modified to present the related sentences found in other abstracts by this automatic procedure.

Abstract Motivation: With the explosion of biomedical literature and the evolution of online and open access, scientists are reading more articles from a wider variety of journals. Thus, the list of core journals relevant to their research may be less obvious and may often change over time. To help researchers quickly identify appropriate journals to read and publish in, we developed a web application for finding related journals based on the analysis of PubMed log data. Availability: http://www.ncbi.nlm.nih.gov/IRET/Journals Contact: [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.

Transaction logs from online search engines are valuable for two reasons: First, they provide insight into human information-seeking behavior. Second, log data can be used to train user models, which can then be applied to improve retrieval systems. This article presents a study of logs from PubMed((R)), the public gateway to the MEDLINE((R)) database of bibliographic records from the medical and biomedical primary literature. Unlike most previous studies on general Web search, our work examines user activities with a highly-specialized search engine. We encode user actions as string sequences and model these sequences using n-gram language models. The models are evaluated in terms of perplexity and in a sequence prediction task. They help us better understand how PubMed users search for information and provide an enabler for improving users' search experience.

This paper investigates the effectiveness of using MeSH(®) in PubMed through its automatic query expansion process: Automatic Term Mapping (ATM). We run Boolean searches based on a collection of 64 topics and about 160,000 MEDLINE(®) citations used in the 2006 and 2007 TREC Genomics Tracks. For each topic, we first automatically construct a query by selecting keywords from the question. Next, each query is expanded by ATM, which assigns different search tags to terms in the query. Three search tags: [MeSH Terms], [Text Words], and [All Fields] are chosen to be studied after expansion because they all make use of the MeSH field of indexed MEDLINE citations. Furthermore, we characterize the two different mechanisms by which the MeSH field is used. Retrieval results using MeSH after expansion are compared to those solely based on the words in MEDLINE title and abstracts. The aggregate retrieval performance is assessed using both F-measure and mean rank precision. Experimental results suggest that query expansion using MeSH in PubMed can generally improve retrieval performance, but the improvement may not affect end PubMed users in realistic situations.

For the purposes of classification it is common to represent a document as a bag of words. Such a representation consists of the individual terms making up the document together with the number of times each term appears in the document. All classification methods make use of the terms. It is common to also make use of the local term frequencies at the price of some added complication in the model. Examples are the naïve Bayes multinomial model (MM), the Dirichlet compound multinomial model (DCM) and the Exponential-Family Approximation of the DCM (EDCM), as well as support vector machines (SVM). Although it is usually claimed that incorporating local word frequency in a document improves text classification performance, we here test whether such claims are true or not. In this paper we show experimentally that simplified forms of the MM, EDCM, and SVM models which ignore the frequency of each word in a document perform about at the same level as MM, DCM, EDCM and SVM models which incorporate local term frequency. We also present a new form of the naïve Bayes multivariate Bernoulli model (MBM) which is able to make use of local term frequency and show again that it offers no significant advantage over the plain MBM. We conclude that word burstiness is so strong that additional occurrences of a word essentially add no useful information to a classifier.

This paper evaluates the retrieval effectiveness of relevance ranking strategies on a collection of 55 queries and about 160,000 MEDLINE((R)) citations used in the 2006 and 2007 Text Retrieval Conference (TREC) Genomics Tracks. The authors study two relevance ranking strategies: term frequency-inverse document frequency (TF-IDF) weighting and sentence-level co-occurrence, and examine their ability to rank retrieved MEDLINE documents given user queries. Furthermore, the authors use the reverse chronological order-PubMed's default display option-as a baseline for comparison. Retrieval effectiveness is assessed using both mean average precision and mean rank precision. Experimental results show that retrievals based on the two strategies had improved performance over the baseline performance, and that TF-IDF weighting is more effective in retrieving relevant documents based on the comparison between the two strategies.

indicates that many such queries are meaningful phrases, rather than simple collections of terms. In this study, we examine whether or not it makes a difference, in terms of retrieval quality, if such queries are interpreted as a phrase or as a conjunction of query terms. And, if it does, what is the optimal way of searching with such queries. To address the question, we developed an automated retrieval evaluation method, based on machine learning techniques, that enables us to evaluate and compare various retrieval outcomes. We show that the class of records that contain all the search terms, but not the phrase, qualitatively differs from the class of records containing the phrase. We also show that the difference is systematic, depending on the proximity of query terms to each other within the record. Based on these results, one can establish the best retrieval order for the records. Our findings are consistent with studies in proximity searching.

Large-scale annotation efforts typically involve several experts who may disagree with each other. We propose an approach for modeling disagreements among experts that allows providing each annotation with a confidence value (i.e., the posterior probability that it is correct). Our approach allows computing certainty-level for individual annotations, given annotator-specific parameters estimated from data. We developed two probabilistic models for performing this analysis, compared these models using computer simulation, and tested each model's actual performance, based on a large data set generated by human annotators specifically for this study. We show that even in the worst-case scenario, when all annotators disagree, our approach allows us to significantly increase the probability of choosing the correct annotation. Along with this publication we make publicly available a corpus of 10,000 sentences annotated according to several cardinal dimensions that we have introduced in earlier work. The 10,000 sentences were all 3-fold annotated by a group of eight experts, while a 1,000-sentence subset was further 5-fold annotated by five new experts. While the presented data represent a specialized curation task, our modeling approach is general; most data annotation studies could benefit from our methodology.

MOTIVATION: Much current research in biomedical text mining is concerned with serving biologists by extracting certain information from scientific text. We note that there is no 'average biologist' client; different users have distinct needs. For instance, as noted in past evaluation efforts (BioCreative, TREC, KDD) database curators are often interested in sentences showing experimental evidence and methods. Conversely, lab scientists searching for known information about a protein may seek facts, typically stated with high confidence. Text-mining systems can target specific end-users and become more effective, if the system can first identify text regions rich in the type of scientific content that is of interest to the user, retrieve documents that have many such regions, and focus on fact extraction from these regions. Here, we study the ability to characterize and classify such text automatically. We have recently introduced a multi-dimensional categorization and annotation scheme, developed to be applicable to a wide variety of biomedical documents and scientific statements, while intended to support specific biomedical retrieval and extraction tasks. RESULTS: The annotation scheme was applied to a large corpus in a controlled effort by eight independent annotators, where three individual annotators independently tagged each sentence. We then trained and tested machine learning classifiers to automatically categorize sentence fragments based on the annotation. We discuss here the issues involved in this task, and present an overview of the results. The latter strongly suggest that automatic annotation along most of the dimensions is highly feasible, and that this new framework for scientific sentence categorization is applicable in practice.

BACKGROUND: The rapid growth of biomedical literature presents challenges for automatic text processing, and one of the challenges is abbreviation identification. The presence of unrecognized abbreviations in text hinders indexing algorithms and adversely affects information retrieval and extraction. Automatic abbreviation definition identification can help resolve these issues. However, abbreviations and their definitions identified by an automatic process are of uncertain validity. Due to the size of databases such as MEDLINE only a small fraction of abbreviation-definition pairs can be examined manually. An automatic way to estimate the accuracy of abbreviation-definition pairs extracted from text is needed. In this paper we propose an abbreviation definition identification algorithm that employs a variety of strategies to identify the most probable abbreviation definition. In addition our algorithm produces an accuracy estimate, pseudo-precision, for each strategy without using a human-judged gold standard. The pseudo-precisions determine the order in which the algorithm applies the strategies in seeking to identify the definition of an abbreviation. RESULTS: On the Medstract corpus our algorithm produced 97% precision and 85% recall which is higher than previously reported results. We also annotated 1250 randomly selected MEDLINE records as a gold standard. On this set we achieved 96.5% precision and 83.2% recall. This compares favourably with the well known Schwartz and Hearst algorithm. CONCLUSION: We developed an algorithm for abbreviation identification that uses a variety of strategies to identify the most probable definition for an abbreviation and also produces an estimated accuracy of the result. This process is purely automatic.

OBJECTIVES: The aim of this study was to improve naïve Bayes prediction of Medical Subject Headings (MeSH) assignment to documents using optimal training sets found by an active learning inspired method. DESIGN: The authors selected 20 MeSH terms whose occurrences cover a range of frequencies. For each MeSH term, they found an optimal training set, a subset of the whole training set. An optimal training set consists of all documents including a given MeSH term (C1 class) and those documents not including a given MeSH term (C(-1) class) that are closest to the C1 class. These small sets were used to predict MeSH assignments in the MEDLINE database. MEASUREMENTS: Average precision was used to compare MeSH assignment using the naïve Bayes learner trained on the whole training set, optimal sets, and random sets. The authors compared 95% lower confidence limits of average precisions of naïve Bayes with upper bounds for average precisions of a K-nearest neighbor (KNN) classifier. RESULTS: For all 20 MeSH assignments, the optimal training sets produced nearly 200% improvement over use of the whole training sets. In 17 of those MeSH assignments, naïve Bayes using optimal training sets was statistically better than a KNN. In 15 of those, optimal training sets performed better than optimized feature selection. Overall naïve Bayes averaged 14% better than a KNN for all 20 MeSH assignments. Using these optimal sets with another classifier, C-modified least squares (CMLS), produced an additional 6% improvement over naïve Bayes. CONCLUSION: Using a smaller optimal training set greatly improved learning with naïve Bayes. The performance is superior to a KNN. The small training set can be used with other sophisticated learning methods, such as CMLS, where using the whole training set would not be feasible.

BACKGROUND: Accurate selection of splice sites during the splicing of precursors to messenger RNA requires both relatively well-characterized signals at the splice sites and auxiliary signals in the adjacent exons and introns. We previously described a feature generation algorithm (FGA) that is capable of achieving high classification accuracy on human 3' splice sites. In this paper, we extend the splice-site prediction to 5' splice sites and explore the generated features for biologically meaningful splicing signals. RESULTS: We present examples from the observed features that correspond to known signals, both core signals (including the branch site and pyrimidine tract) and auxiliary signals (including GGG triplets and exon splicing enhancers). We present evidence that features identified by FGA include splicing signals not found by other methods. CONCLUSION: Our generated features capture known biological signals in the expected sequence interval flanking splice sites. The method can be easily applied to other species and to similar classification problems, such as tissue-specific regulatory elements, polyadenylation sites, promoters, etc.

BACKGROUND: We present a probabilistic topic-based model for content similarity called pmra that underlies the related article search feature in PubMed. Whether or not a document is about a particular topic is computed from term frequencies, modeled as Poisson distributions. Unlike previous probabilistic retrieval models, we do not attempt to estimate relevance-but rather our focus is "relatedness", the probability that a user would want to examine a particular document given known interest in another. We also describe a novel technique for estimating parameters that does not require human relevance judgments; instead, the process is based on the existence of MeSH in MEDLINE. RESULTS: The pmra retrieval model was compared against bm25, a competitive probabilistic model that shares theoretical similarities. Experiments using the test collection from the TREC 2005 genomics track shows a small but statistically significant improvement of pmra over bm25 in terms of precision. CONCLUSION: Our experiments suggest that the pmra model provides an effective ranking algorithm for related article search.

We explore a syntactic approach to sentence compression in the biomedical domain, grounded in the context of result presentation for related article search in the PubMed search engine. By automatically trimming inessential fragments of article titles, a system can effectively display more results in the same amount of space. Our implemented prototype operates by applying a sequence of syntactic trimming rules over the parse trees of article titles. Two separate studies were conducted using a corpus of manually compressed examples from MEDLINE: an automatic evaluation using B leu and a summative evaluation involving human assessors. Experiments show that a syntactic approach to sentence compression is effective in the biomedical domain and that the presentation of compressed article titles supports accurate “interest judgments”, decisions by users as to whether an article is worth examining in more detail.

BACKGROUND: While biomedical text mining is emerging as an important research area, practical results have proven difficult to achieve. We believe that an important first step towards more accurate text-mining lies in the ability to identify and characterize text that satisfies various types of information needs. We report here the results of our inquiry into properties of scientific text that have sufficient generality to transcend the confines of a narrow subject area, while supporting practical mining of text for factual information. Our ultimate goal is to annotate a significant corpus of biomedical text and train machine learning methods to automatically categorize such text along certain dimensions that we have defined. RESULTS: We have identified five qualitative dimensions that we believe characterize a broad range of scientific sentences, and are therefore useful for supporting a general approach to text-mining: focus, polarity, certainty, evidence, and directionality. We define these dimensions and describe the guidelines we have developed for annotating text with regard to them. To examine the effectiveness of the guidelines, twelve annotators independently annotated the same set of 101 sentences that were randomly selected from current biomedical periodicals. Analysis of these annotations shows 70-80% inter-annotator agreement, suggesting that our guidelines indeed present a well-defined, executable and reproducible task. CONCLUSION: We present our guidelines defining a text annotation task, along with annotation results from multiple independently produced annotations, demonstrating the feasibility of the task. The annotation of a very large corpus of documents along these guidelines is currently ongoing. These annotations form the basis for the categorization of text along multiple dimensions, to support viable text mining for experimental results, methodology statements, and other forms of information. We are currently developing machine learning methods, to be trained and tested on the annotated corpus, that would allow for the automatic categorization of biomedical text along the general dimensions that we have presented. The guidelines in full detail, along with annotated examples, are publicly available.

A part-of-speech tagger is a fundamental and indispensable tool in computational linguistics, typically employed at the critical early stages of processing. Although taggers are widely available that achieve high accuracy in very general domains, these do not perform nearly as well when applied to novel specialized domains, and this is especially true with biological text. We present a stochastic tagger that achieves over 97.44% accuracy on MEDLINE abstracts. A primary component of the tagger is its lexicon which enumerates the permitted parts-of-speech for the 10000 words most frequently occurring in MEDLINE. We present evidence for the conclusion that the lexicon is as vital to tagger accuracy as a training corpus, and more important than previously thought.

Abbreviations and acronyms are widely used in the biomedical literature and many of them represent important biomedical concepts. Because many abbreviations are ambiguous (e.g., CAT denotes both chloramphenicol acetyl transferase and computed axial tomography , depending on the context), recognizing the full form associated with each abbreviation is in most cases equivalent to identifying the meaning of the abbreviation. This, in turn, allows us to perform more accurate natural language processing, information extraction, and retrieval. In this study, we have developed supervised approaches to identifying the full forms of ambiguous abbreviations within the context they appear. We first automatically assigned multiple possible full forms for each abbreviation; we then treated the in-context full-form prediction for each specific abbreviation occurrence as a case of word-sense disambiguation. We generated automatically a dictionary of all possible full forms for each abbreviation. We applied supervised machine-learning algorithms for disambiguation. Because some of the links between abbreviations and their corresponding full forms are explicitly given in the text and can be recovered automatically, we can use these explicit links to automatically provide training data for disambiguating the abbreviations that are not linked to a full form within a text. We evaluated our methods on over 150 thousand abstracts and obtain for coverage and precision results of 82% and 92%, respectively, when performed as tenfold cross-validation, and 79% and 80%, respectively, when evaluated against an external set of abstracts in which the abbreviations are not defined.

BACKGROUND: Named entity recognition (NER) is an important first step for text mining the biomedical literature. Evaluating the performance of biomedical NER systems is impossible without a standardized test corpus. The annotation of such a corpus for gene/protein name NER is a difficult process due to the complexity of gene/protein names. We describe the construction and annotation of GENETAG, a corpus of 20K MEDLINE sentences for gene/protein NER. 15K GENETAG sentences were used for the BioCreAtIvE Task 1A Competition. RESULTS: To ensure heterogeneity of the corpus, MEDLINE sentences were first scored for term similarity to documents with known gene names, and 10K high- and 10K low-scoring sentences were chosen at random. The original 20K sentences were run through a gene/protein name tagger, and the results were modified manually to reflect a wide definition of gene/protein names subject to a specificity constraint, a rule that required the tagged entities to refer to specific entities. Each sentence in GENETAG was annotated with acceptable alternatives to the gene/protein names it contained, allowing for partial matching with semantic constraints. Semantic constraints are rules requiring the tagged entity to contain its true meaning in the sentence context. Application of these constraints results in a more meaningful measure of the performance of an NER system than unrestricted partial matching. CONCLUSION: The annotation of GENETAG required intricate manual judgments by annotators which hindered tagging consistency. The data were pre-segmented into words, to provide indices supporting comparison of system responses to the "gold standard". However, character-based indices would have been more robust than word-based indices. GENETAG Train, Test and Round1 data and ancillary programs are freely available at ftp://ftp.ncbi.nlm.nih.gov/pub/tanabe/GENETAG.tar.gz. A newer version of GENETAG-05, will be released later this year.

Biomedical abbreviations and acronyms are widely used in biomedical literature. Since many abbreviations represent important content in biomedical literature, information retrieval and extraction benefits from identifying the meanings of biomedical abbreviations. Since many abbreviations are ambiguous, it would be important to map abbreviations to their full forms, which ultimately represent the meanings of the abbreviations. In this study, we present a novel unsupervised method that applies MEDLINE records as a knowledge source for disambiguating abbreviations in full-text biomedical journal articles. We first automatically generated from MEDLINE records a knowledge source or dictionary of abbreviation-full pairs. We then trained on MEDLINE records and predicted the full forms of abbreviations in full-text journal articles by applying supervised machine-learning algorithms in an unsupervised fashion. We report up to 92% prediction precision and up to 91% coverage.

SUMMARY: We present a part-of-speech tagger that achieves over 97% accuracy on MEDLINE citations. AVAILABILITY: Software, documentation and a corpus of 5700 manually tagged sentences are available at ftp://ftp.ncbi.nlm.nih.gov/pub/lsmith/MedPost/medpost.tar.gz

Abstract MEDLINE® is a collection of more than 12 million references and abstracts covering recent life science literature. With its continued growth and cutting‐edge terminology, spell‐checking with a traditional lexicon based approach requires significant additional manual follow‐up. In this work, an internal corpus based context quality rating α, frequency, and simple misspelling transformations are used to rank words from most likely to be misspellings to least likely. Eleven‐point average precisions of 0.891 have been achieved within a class of 42,340 all alphabetic words having an α score less than 10. Our models predict that 16,274 or 38% of these words are misspellings. Based on test data, this result has a recall of 79% and a precision of 86%. In other words, spell checking can be done by statistics instead of with a dictionary. As an application we examine the time history of low α words in MEDLINE® titles and abstracts.

MOTIVATION: The MEDLINE database of biomedical abstracts contains scientific knowledge about thousands of interacting genes and proteins. Automated text processing can aid in the comprehension and synthesis of this valuable information. The fundamental task of identifying gene and protein names is a necessary first step towards making full use of the information encoded in biomedical text. This remains a challenging task due to the irregularities and ambiguities in gene and protein nomenclature. We propose to approach the detection of gene and protein names in scientific abstracts as part-of-speech tagging, the most basic form of linguistic corpus annotation. RESULTS: We present a method for tagging gene and protein names in biomedical text using a combination of statistical and knowledge-based strategies. This method incorporates automatically generated rules from a transformation-based part-of-speech tagger, and manually generated rules from morphological clues, low frequency trigrams, indicator terms, suffixes and part-of-speech information. Results of an experiment on a test corpus of 56K MEDLINE documents demonstrate that our method to extract gene and protein names can be applied to large sets of MEDLINE abstracts, without the need for special conditions or human experts to predetermine relevant subsets. AVAILABILITY: The programs are available on request from the authors.

An important problem in the indexing of natural language text is how to identify those words and phrases that reflect the content of the text. In general, automatic indexing has dealt with this problem by removing instances of a few hundred common words known as stop words, and treating the remaining words as though they were content bearing. This approach is acceptable for some applications such as statistical estimates of the similarity of queries and documents for the purpose of document retrieval. However, when the indexing terms are to be examined by a human as a means of accessing the literature, it greatly improves efficiency if most of the noncontent-bearing words and phrases can be eliminated from the indexing. Here we present three statistical techniques for identifying content-bearing phrases within a natural language database. We demonstrate the effectiveness of the methods on test data, and show how all three methods can be combined to produce a single improved method.

PURPOSE: The authors study the extraction of useful phrases from a natural language database by statistical methods. The aim is to leverage human effort by providing preprocessed phrase lists with a high percentage of useful material. METHOD: The approach is to develop six different scoring methods that are based on different aspects of phrase occurrence. The emphasis here is not on lexical information or syntactic structure but rather on the statistical properties of word pairs and triples that can be obtained from a large database. MEASUREMENTS: The Unified Medical Language System (UMLS) incorporates a large list of humanly acceptable phrases in the medical field as a part of its structure. The authors use this list of phrases as a gold standard for validating their methods. A good method is one that ranks the UMLS phrases high among all phrases studied. Measurements are 11-point average precision values and precision-recall curves based on the rankings. RESULT: The authors find of six different scoring methods that each proves effective in identifying UMLS quality phrases in a large subset of MEDLINE. These methods are applicable both to word pairs and word triples. All six methods are optimally combined to produce composite scoring methods that are more effective than any single method. The quality of the composite methods appears sufficient to support the automatic placement of hyperlinks in text at the site of highly ranked phrases. CONCLUSION: Statistical scoring methods provide a promising approach to the extraction of useful phrases from a natural language database for the purpose of indexing or providing hyperlinks in text.

Useful retrieval depends on the ability to predict which documents a user will find helpful in answer to a query. Our interest is the common case when no information is provided about the user other than the query and the query is in natural language. In this setting it is well accepted that a human can make useful predictions in the form of judgments about what will likely prove useful to another human. We present data showing that when the predictions of a group of humans are averaged, the result is a better predictor. If performance is measured as a precision, the group performance increases with the size of the group and approaches a limit of approximately 50% improvement over average individual performance on our data. Superior performance by groups raises the question of how. The groups we studied were subject experts and a natural question was whether the superior performance resulted from the pooling of their subject knowledge. In order to answer this question we studied also a group of untrained individuals. To our surprise we found that while untrained individuals had a somewhat inferior performance compared to trained individuals, the group of untrained individuals together performed better than any single trained individual and almost at the level of the trained group. © 1998 John Wiley & Sons, Inc.

We show first that the pooling of multiple human judgments of relevance provides predictor of relevance that is superior to that obtained from a single human's relevance judgemts. A learning algorithm applied to a set of relevance judgments obtained from a single human would be expected to perform on new material at a level somewhat below that human. However, we examine two learning methods which when trained on the superior source of pooled human relevance judgments are able to perform at the level of a single human on new material. All performance comparisons are based on an independent human judge. Both algorithms function by producing term weights—one by a log odds calculation and the other by producing a least-squares fit to human relevance ratings. Some characteristics of the algorithms are examined.

This article studies aggressive word removal in text categorization to reduce the noise in free texts and to enhance the computational efficiency of categorization. We use a novel stop word identification method to automatically generate domain specific stoplists which are much larger than a conventional domain-independent stoplist. In our tests with three categorization methods on text collections from different domains/applications, significant numbers of words were removed without sacrificing categorization effectiveness. In the test of the Expert Network method on CACM documents, for example, an 87% removal of unique words reduced the vocabulary of documents from 8,002 distinct words to 1,045 words, which resulted in a 63% time savings and a 74% memory savings in the computation of category ranking, with a 10% precision improvement on average over not using word removal. It is evident in this study that automated word removal based on corpus statistics has a practical and significant impact on the computational tractability of categorization methods in large databases. © 1996 John Wiley & Sons, Inc.

If one could identify the source subject areas of documents and could compute the probability that any given document came from a given source, one could apply Baye's theorem to compute the probability that a query document and any other document came from the same subject area (i.e., were related). Even correct prior probabilities could be assigned under this hypothesis by examining the whole database to obtain the probabilities with which different sources occur. While we do not know how to carry out this scheme in such a way as to account for all the information contained in documents, we show here how it may be realized in a limited way. A method of modeling the sources of documents is described which accounts for the information in global term weights. The methodology is based on the hypergeometric probability distribution. Such a source model may be fit closely to a real database and may be used to convert the real database to an abstract database in which document sources are known and model retrieval is the best retrieval possible based on model document content. We have constructed such an abstract model corresponding to a database of MEDLINE records. Tests of vector retrieval methods on the abstract model indicate they are near optimal but suggest minor improvement with correct parameter choices. Preliminary results based on a test set (human judged) from the real database support these results. © 1993 John Wiley & Sons, Inc.

A procedure for retrieval testing is developed which is based on the comparison of statistically independent methods of retrieval applied to the same database. This is, in essence, a form of relative retrieval testing in which, due to the independence assumption, the conclusions drawn may be taken as absolute. The methodology is illustrated on a large database of MEDLINE records where documents are themselves used as queries. Test results in this setting indicate that, among the several methods tested, a form of cosine coefficient retrieval will give the best performance for the purpose of clustering documents or producing nearest neighbors of documents. © 1992 John Wiley & Sons, Inc.