Personalized Retrieval-Augmented Generation (RAG) relies on accurately selecting userrelevant documents.In practice, existing RAG approaches often suffer from high retrieval costs and overlook that collaborative signals from similar users can enhance personalized generation for the current user.We propose ClusterRAG, a Cluster-Based Collaborative Filtering for Personalized Retrieval-Augmented Generation.ClusterRAG represents users through their profile documents, organizes users into semantically coherent clusters using density-based clustering, and performs retrieval at both the cluster and document levels via cluster-level similarity and finegrained ranking.Extensive experiments on the LaMP benchmark demonstrate that jointly leveraging the target user's profile and profiles from top similar users consistently yields the best performance across diverse tasks.Further analysis shows that ClusterRAG integrates seamlessly with different dense retrievers and rankers, and remains effective when paired with both fine-tuned and zero-shot language models.

NLP models often rely on human-labeled data for training and evaluation.Many approaches crowdsource this data from a large number of annotators with varying skills, backgrounds, and motivations, resulting in conflicting annotations.These conflicts have traditionally been resolved by aggregation methods that assume disagreements are errors.Recent work has argued that for many tasks annotators may have genuine disagreements and that variation should be treated as signal rather than noise.However, few models separate signal and noise in annotator disagreement.In this work, we introduce NUTMEG, a new Bayesian model that incorporates information about annotator backgrounds to remove noisy annotations from human-labeled training data while preserving systematic disagreements.Using synthetic and real-world data, we show that NUTMEG is more effective at recovering ground-truth from annotations with systematic disagreement than traditional aggregation methods, and we demonstrate that downstream models trained on NUT-MEG-aggregated data significantly outperform models trained on data from traditionally aggregation methods.We provide further analysis characterizing how differences in subpopulation sizes, rates of disagreement, and rates of spam affect the performance of our model.Our results highlight the importance of accounting for both annotator competence and systematic disagreements when training on human-labeled data.

Unsupervised Domain Adaptation has been an efficient approach to transferring the semantic segmentation model across data distributions. Meanwhile, the recent Open-vocabulary Semantic Scene understanding based on large-scale vision language models is effective in open-set settings because it can learn diverse concepts and categories. However, these prior methods fail to generalize across different camera views due to the lack of cross-view geometric modeling. At present, there are limited studies analyzing cross-view learning. To address this problem, we introduce a novel Unsupervised Cross-view Adaptation Learning approach to modeling the geometric structural change across views in Semantic Scene Understanding. First, we introduce a novel Cross-view Geometric Constraint on Unpaired Data to model structural changes in images and segmentation masks across cameras. Second, we present a new Geodesic Flow-based Correlation Metric to efficiently measure the geometric structural changes across camera views. Third, we introduce a novel view-condition prompting mechanism to enhance the view-information modeling of the open-vocabulary segmentation network in cross-view adaptation learning. The experiments on different cross-view adaptation benchmarks have shown the effectiveness of our approach in cross-view modeling, demonstrating that we achieve State-of-the-Art (SOTA) performance compared to prior unsupervised domain adaptation and open-vocabulary semantic segmentation methods.

This manuscript describes the development of a resources module that is part of a learning platform named 'NIGMS Sandbox for Cloud-based Learning' https://github.com/NIGMS/NIGMS-Sandbox. The overall genesis of the Sandbox is described in the editorial NIGMS Sandbox at the beginning of this Supplement. This module delivers learning materials on implementing deep learning algorithms for biomedical image data in an interactive format that uses appropriate cloud resources for data access and analyses. Biomedical-related datasets are widely used in both research and clinical settings, but the ability for professionally trained clinicians and researchers to interpret datasets becomes difficult as the size and breadth of these datasets increases. Artificial intelligence, and specifically deep learning neural networks, have recently become an important tool in novel biomedical research. However, use is limited due to their computational requirements and confusion regarding different neural network architectures. The goal of this learning module is to introduce types of deep learning neural networks and cover practices that are commonly used in biomedical research. This module is subdivided into four submodules that cover classification, augmentation, segmentation and regression. Each complementary submodule was written on the Google Cloud Platform and contains detailed code and explanations, as well as quizzes and challenges to facilitate user training. Overall, the goal of this learning module is to enable users to identify and integrate the correct type of neural network with their data while highlighting the ease-of-use of cloud computing for implementing neural networks. This manuscript describes the development of a resource module that is part of a learning platform named ``NIGMS Sandbox for Cloud-based Learning'' https://github.com/NIGMS/NIGMS-Sandbox. The overall genesis of the Sandbox is described in the editorial NIGMS Sandbox [1] at the beginning of this Supplement. This module delivers learning materials on the analysis of bulk and single-cell ATAC-seq data in an interactive format that uses appropriate cloud resources for data access and analyses.

Micro-expression recognition is one of the most challenging topics in affective computing. It aims to recognize tiny facial movements difficult for humans to perceive in a brief period, i.e., 0.25 to 0.5 seconds. Recent advances in pre-training deep Bidirectional Transformers (BERT) have significantly improved self-supervised learning tasks in computer vision. However, the standard BERT in vision problems is designed to learn only from full images or videos, and the architecture cannot accurately detect details of facial micro-expressions. This paper presents Micron-BERT ($(\mu$-BERT), a novel approach to facial micro-expression recognition. The proposed method can automatically capture these movements in an unsupervised manner based on two key ideas. First, we employ Diagonal Micro-Attention (DMA) to detect tiny differences between two frames. Second, we introduce a new Patch of Interest (PoI) module to localize and highlight micro-expression interest regions and simultaneously reduce noisy backgrounds and distractions. By incorporating these components into an end-to-end deep network, the proposed$\mu$-BERT significantly outperforms all previous work in various micro-expression tasks.$\mu$-BERT can be trained on a large-scale unlabeled dataset, i.e., up to 8 million images, and achieves high accuracy on new unseen facial micro-expression datasets. Empirical experiments show$\mu$-BERT consistently outperforms state-of-the-art performance on four micro-expression benchmarks, including SAMM, CASME II, SMIC, and CASME3, by significant margins. Code will be available at https://github.com/uark-cviu/Micron-BERT

Linked Open Data (LOD) has been exploited by recommender systems in different ways. One way measures the semantic distance between LOD resources so that nearby resources are considered related and can be recommended, the LDSD method exemplifies this approach. One drawback of the LDSD approach is that it can only compute the semantic distance between two resources that are directly linked or indirectly linked through an intermediate resource. Resources that are located more than two links away are automatically considered unrelated. In this paper, we introduce an approach, called Propagated Linked Data Semantic Distance (PLDSD), that expands the coverage of current semantic distance approaches. We employ an all-pair shortest path algorithm, the well-known Floyd-Warshall algorithm, to expand semantic distance calculations beyond resources that are just one or two links away. To validate the effectiveness of our approach, we conducted an experiment to identify the relatedness between musical artists in DBpedia, and it demonstrated that, by increasing the reach of semantic distance calculations, PLDSD increases the accuracy of the LOD-based recommendations over our baselines.

The CiteSeerx project at the University of Arkansas uses a browsing interface is based on the Association for Computing Machinery's Computing Classification System (ACM CCS). CCS contains just 369 categories whereas the CiteSeerx database contains over 2 million documents. This results in more than 6500 documents per category, far too many to browse. To address this problem, we are exploring ways to automatically expand the CCS ontology. Previous work has focused on using clustering to automatically identify the new classes. This work focuses on how to label the subclasses in a semantically meaningful way to that they can support user browsing. We develop methods based on text mining from the subclass members to extract class labels. We evaluate three methods by comparing the suggested labels with human-assigned labels for existing categories.

This paper presents a new services paradigm for virtual world crawler interaction that is co-operative and exploits information about 3D objects in the virtual world. Our approach supports analyzing redundant information crawled from virtual worlds in order to decrease the amount of data collected by crawlers, keep search engine collections up to date, and provide an efficient mechanism for collecting and searching information from multiple virtual worlds.

An important task of ontology learning is to enrich the vocabulary for domain ontologies using different sources of information. WordNet, an online lexical database covering many domains, has been widely used as a source from which to mine new vocabulary for ontology enrichment. However, since each word submitted to WordNet may have several different meanings (senses), existing approaches still face the problem of semantic disambiguation in order to select the correct sense for the new vocabulary to be added. In this paper, we present a similarity computation method that allows us to efficiently select the correct WordNet sense for a concept-word in a given ontology. Once the correct sense is identified, we can then enrich the concept's vocabularly using nearby words in WordNet. Experimental results using an amphibian ontology show that the similarity computation method reach a good average accuracy and our approach is able to enrich the vocabulary of each concept with words mined from WordNet synonyms and hypernyms.

Short search engine queries do not provide contextual information, making it difficult for traditional search engines to understand what users are really requesting. One approach to this problem is to use recommender systems that identify user interests through various methods in order to provide information specific to the user's needs. However, many current recommender systems use a collaborative model based on a network of users to provide the recommendations, leading to problems in environments where network relationships are sparse or unknown. Content-based recommenders can avoid the sparsity problem but they may be inefficient for large document collections. In this paper, we propose a concept-based recommender system that recommends papers to general users of the CiteSeerx digital library of Computer Science research publications. We also represent a novel way of classifying documents and creating user profiles based on the ACM (Association for Computer Machinery) classification tree. Based on these user profiles which are built using past click histories, relevant papers in the domain are recommended to users. Experiments with a set of users on the CiteSeerX database show that our concept-based method provides accurate recommendations even with limited user profile histories.

We present a system for collecting content from 3D multi-user virtual worlds for use in a cross-world search engine, an enabling technology for linking virtual worlds to the wider web. We use an intelligent agent crawler designed to collect user-generated content without relying on access to private internal server databases. The agents navigate autonomously through the world and interact with content to discover regions, parcels of land within regions, user-created objects, other avatars, and user associations. The experiments we performed are the first which focus on the content within a large virtual world. Our results show that virtual worlds can be effectively crawled using autonomous agent crawlers that emulate normal user behavior. Additionally, we find that the collection of interactive content enhances our ability to identify dynamic, immersive environments within the world.

In this paper, we proposed a new generalized Multivariate Probalistic Modeling (MPM) to automatically extract topics from text collection and attach them with existing ontology. Specially, we first make use of KeyConcept which is a classification system classify documents into a set of predefined concepts. Then, by modeling documents cluster based MPM, we extract latent concepts and corrensponding sub-clusters from document collection. We compare our MPM with Probabilistic Latent Semantic Indexing (PLSI) and other clustering algorithm on Citeseerx data sets. Experiment results show that MPM outperforms PLSI in terms of time efficiency and provides better topics representation. Clustering analysis also prove the advantages of our MPM over other clustering technique in precision.

Abstract Text classifiers automatically classify documents into appropriate concepts for different applications. Most classification approaches use flat classifiers that treat each concept as independent, even when the concept space is hierarchically structured. In contrast, hierarchical text classification exploits the structural relationships between the concepts. In this article, we explore the effectiveness of hierarchical classification for a large concept hierarchy. Since the quality of the classification is dependent on the quality and quantity of the training data, we evaluate the use of documents selected from subconcepts to address the sparseness of training data for the top‐level classifiers and the use of document relationships to identify the most representative training documents. By selecting training documents using structural and similarity relationships, we achieve a statistically significant improvement of 39.8% (from 54.5–76.2%) in the accuracy of the hierarchical classifier over that of the flat classifier for a large, three‐level concept hierarchy.

Users organize personal information in various ways. We believe that this process could be expedited and improved by using domain ontologies. The main problem with this idea is the lack of automatic tools that help non-expert users to build and maintain their own ontologies. In this study we report progress in the process of adapting ontologies to better represent a given set of documents centered on a topic of interest. More specifically we investigate automatic approaches to enhance the representation of the concepts within the domain ontology. We show that our approach can enrich the vocabulary of each concept with words mined from the set of small documents provided. The method we propose is based on efficient text mining approaches combined with semantic information from WordNet.

The increasing interest in the Semantic Web is producing a growing number of publicly available domain ontologies. These ontologies are a rich source of information that could be very helpful during the process of engineering other domain ontologies. We present an automatic technique that, given a set of Web documents, selects appropriate domain ontologies from a collection of pre-existing ontologies. We empirically compare an ontology match score that is based on statistical techniques with simple keyword matching algorithms. The algorithms were tested on a set of 183 publicly available ontologies and documents representing ten different domains. Our algorithm was able to select the correct domain ontology as the top ranked ontology 8 out of 10 times.

User profiles, descriptions of user interests, can be used by search engines to provide personalized search results. Many approaches to creating user profiles collect user information through proxy servers (to capture browsing histories) or desktop bots (to capture activities on a personal computer). Both these techniques require participation of the user to install the proxy server or the bot. In this study, we explore the use of a less-invasive means of gathering user information for personalized search. In particular, we build user profiles based on activity at the search site itself and study the use of these profiles to provide personalized search results. By implementing a wrapper around the Google search engine, we were able to collect information about individual user search activities. In particular, we collected the queries for which at least one search result was examined, and the snippets (titles and summaries) for each examined result. User profiles were created by classifying the collected information (queries or snippets) into concepts in a reference concept hierarchy. These profiles were then used to re-rank the search results and the rank-order of the user-examined results before and after re-ranking were compared. Our study found that user profiles based on queries were as effective as those based on snippets. We also found that our personalized re-ranking resulted in a 34% improvement in the rankorder of the user-selected results.

User profiles, descriptions of use interests, can be used by search engines to provide personalized search results. Many approaches to creating user profiles collect user information through proxy servers (to capture browsing histories) or desktop bots (to capture activities on a personal computer). Both these techniques require participation of the user to install the proxy server or bot. In a previous study [1], we explored the use of a less-invasive means of gathering user information for personalized search. In particular, we built user profiles based on activity at the search site itself and studied the use of these profiles to provide personalized search results. By implementing a wrapper around the Google search engine, we were able to collect information about individual user search activities. More specifically, we collected the queries for which at least one search result was examined, and the snippets (titles and summaries) for each examined result. User profiles were created by classifying the collected information (queries or snippets) into concepts in a reference concept hierarchy. These profiles were then used to re-rank the search results and the rank-order of the user-examined results before and after re-ranking were compared. Our study found that user profiles based on queries were as effective as those based on snippets. We also found that our personalized re-ranking resulted in a 34% improvement in the rank-order of the user-selected results. In this demo we present misearch (http://moby.ittc.ku.edu/⁓micro/demo), a meta-search engine that implemented the algorithm analyzed in the previous study.

As the number of available Web pages grows, users experience increasing difficulty finding documents relevant to their interests. One of the underlying reasons for this is that most search engines find matches based on keywords, regardless of their meanings. To provide the user with more useful information, we need a system that disambiguates queries by including information about the user's conceptual framework. This is the goal of KeyConcept, a conceptual search engine. During indexing, KeyConcept classifies documents into concepts selected from a manuallyconstructed concept hierarchy. During retrieval, KeyConcept ranks documents based on a combination of keyword and conceptual similarity. This paper describes the system architecture and discusses the results of experiments that evaluate the effect of exploiting the hierarchical relationships between concepts during retrieval. Our results confirm that conceptual match significantly improves the precision of the search results over keyword match alone. In addition, the use of the concept hierarchy to prune irrelevant search results also significantly increases precision. Finally, we show that the two used together, conceptual search and pruning, significantly outperforms either used alone.

The publicly indexable Web contains an estimated 800 million pages, however it is estimated that the largest search engine contains only 300 million of these pages.As the number of Internet users and the number of accessible Web pages grows, it is becoming increasingly difficult for users to find documents that are relevant to their particular needs.Often users must browse through a large hierarchy of categories to find the information for which they are looking.To provide the user with the most useful information in the least amount of time, we need a system that uses each user's view of the world for classification.This paper explores a way to use a user's personal arrangement of concepts to navigate the Web.This system is built by using the characterizations for a particular site created by the Ontology Based Informing Web Agent Navigation (OBIWAN) system and mapping from them to the user's personal ontologies.OBIWAN allows users to explore multiple sites via the same browsing hierarchy.This paper extends OBIWAN to allow users to explore multiple sites via their own browsing hierarchy.The mapping of the reference ontology to the personal ontology is shown to have a promising level of correctness and precision.

Most information retrieval systems on the Internet rely primarily on similarity ranking algorithms based solely on term frequency statistics. Information quality is usually ignored. This leads to the problem that documents are retrieved without regard to their quality. We present an approach that combines similarity-based similarity ranking with quality ranking in centralized and distributed search environments. Six quality metrics, including the currency, availability, information-to-noise ratio, authority, popularity, and cohesiveness, were investigated. Search effectiveness was significantly improved when the currency, availability, information-to-noise ratio and page cohesiveness metrics were incorporated in centralized search. The improvement seen when the availability, information-to- noise ratio, popularity, and cohesiveness metrics were incorporated in site selection was also significant. Finally, incorporating the popularity metric in information fusion resulted in a significant improvement. In summary, the results show that incorporating quality metrics can generally improve search effectiveness in both centralized and distributed search environments.

Centralized search process requires that the whole collection reside at a single site. This imposes a burden on both the system storage of the site and the network traffic near the site. It thus comes to require the search process to be distributed. Recently, more and more Web sites provide the ability to search their local collection of Web pages. Query brokering systems are used to direct queries to the promising sites and merge the results from these sites. Creation of meta-information of the sites plays an important role in such systems. In this article, we introduce an ontology-based web site mapping method used to produce conceptual meta-information, the Vector Space approach, and present a serial of experiments comparing it with Naïve-Bayes approach. We found that the Vector Space approach produces better accuracy in ontology-based web site mapping.

With the exponentially growing amount of information available on the Internet, the task of retrieving documents of interest has become increasingly difficult. Search engines usually return more than 1,500 results per query, yet out of the top twenty results, only one half turn out to be relevant to the user. One reason for this is that Web queries are in general very short and give an incomplete specification of individual users' information needs. This paper explores ways of incorporating users' interests into the search process to improve the results. The user profiles are structured as a concept hierarchy of 4,400 nodes. These are populated by 'watching over a user's shoulder' while he is surfing. No explicit feedback is necessary. The profiles are shown to converge and to reflect the actual interests quite well. One possible deployment of the profiles is investigated: re-ranking and filtering search results. Increases in performance are moderate but noticeable and show that fully automatic creation of large hierarchical user profiles is possible.

Searching online text collections can be both rewarding and frustrating. While valuable information can be found, typically many irrelevant documents are also retrieved, while many relevant ones are missed. Terminology mismatches between the user's query and document contents are a main cause of retrieval failures. Expanding a user's query with related words can improve search performances, but finding and using related words is an open problem. This research uses corpus analysis techniques to automatically discover similar words directly from the contents of the databases which are not tagged with part-of-speech labels. Using these similarities, user queries are automatically expanded, resulting in conceptual retrieval rather than requiring exact word matches between queries and documents. We are able to achieve a 7.6% improvement for TREC 5 queries and up to a 28.5% improvement on the narrow-domain Cystic Fibrosis collection. This work has been extended to multidatabase collections where each subdatabase has a collection-specific similarity matrix associated with it. If the best matrix is selected, substantial search improvements are possible. Various techniques to select the appropriate matrix for a particular query are analyzed, and a 4.8% improvement in the results is validated.

Article Associated biological information retrieval from distributed databases Share on Authors: Mousheng Xu Department of Electrical Engineering and Computer Science, University of kansas, Lawrence, KS Department of Electrical Engineering and Computer Science, University of kansas, Lawrence, KSView Profile , Susan Gauch Department of Electrical Engineering and Computer Science, University of kansas, Lawrence, KS Department of Electrical Engineering and Computer Science, University of kansas, Lawrence, KSView Profile Authors Info & Claims CIKM '98: Proceedings of the seventh international conference on Information and knowledge managementNovember 1998 Pages 193–200https://doi.org/10.1145/288627.288657Online:01 November 1998Publication History 1citation725DownloadsMetricsTotal Citations1Total Downloads725Last 12 Months4Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

Inexperienced users typically obtain one of three possible outcomes when they search for online information: They are buried under an information avalanche, they are unable to locate any useful information at all, or they find what they need in roughly the amount they need. Unfortunately, the latter outcome is the most rare. Unfamiliarity with search tactics creates difficulties for many users of online retrieval systems. When faced with poor results, even experienced searchers may use vocabulary incorrectly and often fail to reformulate their queries. Far from being the answer to everyone's information dreams, distributed sources of online information, i.e., the World Wide Web (WWW), compound the problem and may often turn into an information nightmare. To address this problem, intelligent online search assistants, or agents, are being developed for information retrieval applications. There are many approaches, both theoretical and implemented, to using intelligent software agents for information retrieval purposes. These approaches range from desktop agents specialized for a single user to networks of agents used to collect data from distributed information sources, including Web sites. This article presents an overview of intelligent software agents in information retrieval, including an explanation of agents and agent architectures, and presents several agent systems. We distinguish between agents as individual entities, whose properties and characteristics we describe separately, and agent systems as collections of agents utilized for information retrieval tasks, which we discuss in terms of individual implementations. © 1998 John Wiley & Sons, Inc.

Article Free Access Share on A corpus analysis approach for automatic query expansion Authors: Susan Gauch Electical Engineering and Computer Science, 415 Snow Hall, University of Kansas, Lawrence, KS Electical Engineering and Computer Science, 415 Snow Hall, University of Kansas, Lawrence, KSView Profile , Jianying Wang Electical Engineering and Computer Science, 415 Snow Hall, University of Kansas, Lawrence, KS Electical Engineering and Computer Science, 415 Snow Hall, University of Kansas, Lawrence, KSView Profile Authors Info & Claims CIKM '97: Proceedings of the sixth international conference on Information and knowledge managementJanuary 1997 Pages 278–284https://doi.org/10.1145/266714.266910Published:01 January 1997Publication History 17citation423DownloadsMetricsTotal Citations17Total Downloads423Last 12 Months11Last 6 weeks4 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF

Unfamiliarity with search tactics creates difficulties for many users of online retrieval systems. User observations indicate that even experienced searchers use vocabulary incorrectly and rarely reformulate their queries. To address these problems, an expert system for online search assistance was developed. This prototype automatically reformulates queries to improve the search results, and ranks the retrieved passages to speed the identification of relevant information. Users' search performance using the expert system was compared with their search performance on their own, and their search performance using an online thesaurus. The following conclusions were reached: (1) The expert system significantly reduced the number of queries necessary to find relevant passages compared with the user searching alone or with the thesaurus. (2) The expert system produced marginally significant improvements in precision compared with the user searching on their own. There was no significant difference in the recall achieved by the three system configurations. (3) Overall, the expert system ranked relevant passages above irrelevant passages. © 1993 John Wiley & Sons, Inc.

Researchers are exploring the application of artificial intelligence techniques to information retrieval with the goal of providing intelligent access to online information. This article surveys several such systems to show what is possible in the lab today, and what may be possible in the library or office of tomorrow. Systems incorporating user modeling, natural language understanding, and expert systems technology are presented. © 1992 John Wiley & Sons, Inc.

Users of online retrievalsystems experience many difficulties, particularly with search tactics, User studies have indicated that searchers use vocabulary incorrectly and do not take full advantage of iteration to improve their queries.To address these problems, an expert system for online search assistance was developed.This prototype augments the searching capabilities of novice users by providing automatic query reformulation to improve the search results, and automatic ranking of the retrieved passages to speed the identification of relevant information.Users' search performance using the expert system was compared with their search performance on their own, and their search performance using an online thesaurus.The following conclusions were reached: (1) the expert system significantly reduced the number of queries necessary to find relevant passages compared with the user searching alone or with the thesaurus.(2) The expert system produced marginally significant improvements in precision compared with the user searching on their own.There was no significant difference in the recall achieved by the three system configurations.(3) Overall, the expert system ranked relevant passages above irrelevant passages.

This paper presents a prototype expert system which provides online search assistance. The expert system automatically reformulates queries, using an online thesaurus as the source of domain knowledge, and a knowledge base of domain-independent search tactics. The expert system works with a full-text database which requires no syntactic or semantic pre-processing. In addition, the expert system ranks the retrieved passages in decreasing order of probable relevance.