Understanding customer requirements is a key success factor for both business-to-consumer (B2C) and business-to-business (B2B) enterprises. In a B2C context, most requirements are directly related to products and therefore expressed in keyword-based queries. In comparison, B2B requirements contain more information about customer needs and as such the queries are often in a longer form. Such long-form queries pose significant challenges to the information retrieval task in B2B context. In this work, we address the long-form information retrieval challenges by proposing a combination of (i) traditional retrieval methods, to leverage the lexical match from the query, and (ii) state-of-the-art sentence transformers, to capture the rich context in the long queries. We compare our method against traditional TF-IDF and BM25 models on an internal dataset of 12,368 pairs of long-form requirements and products sold. The evaluation shows promising results and provides directions for future work.

Recent advances in text representation have shown that training on large amounts of text is crucial for natural language understanding. However, models trained without predefined notions of topical interest typically require careful fine-tuning when transferred to specialized domains. When a sufficient amount of within-domain text may not be available, expanding a seed corpus of relevant documents from large-scale web data poses several challenges. First, corpus expansion requires scoring and ranking each document in the collection, an operation that can quickly become computationally expensive as the web corpora size grows. Relying on dense vector spaces and pairwise similarity adds to the computational expense. Secondly, as the domain concept becomes more nuanced, capturing the long tail of domain-specific rare terms becomes non-trivial, especially under limited seed corpora scenarios.

The use of AI in knowledge dense domains, e.g., chemistry, medicine, biology, etc. - is extremely promising, but often suffers from slow deployment and adaptation to different tasks. We propose a methodology to quickly capture the intent and expertise of a domain expert in order to train personalized AI models for specific tasks. Specifically we focus on the domain of polymer materials design and discovery: it often takes 10 years or more to design, synthesize, test, and introduce a new polymer material into the market. One way to accelerate up the design of polymer materials is through the use of computational methods to design the material, such as combinatorial screening, generative models, inverse design, etc. The drawback of these methods is that they generate a large number of candidates for new molecules, which then need to be manually reviewed by subject matter experts who select only a dozen for further investigation. Our solution is a human-in-the-loop methodology where we rank the candidates according to a utility function that is learned via the continued interaction with the subject matter experts, but which is also constrained by specific chemical knowledge. We prove the viability of our proposed methodology in a polymer production lab and we (i) evaluate against datasets of polymers previously produced in the lab as well as (ii) producing several novel materials that are undergoing experimental development, and (iii) quantitatively show that standard synthetic accessibility scores do not inform about patterns of SME decisions.

Many Knowledge Extraction systems rely on semantic resources - dictionaries, ontologies, lexical resources - to extract information from unstructured text. A key for successful information extraction is to consider such resources as evolving artifacts and keep them up-to-date. In this paper, we tackle the problem of dictionary expansion and we propose a human-in-the-loop approach: we couple neural language models with tight human supervision to assist the user in building and maintaining domain-specific dictionaries. The approach works on any given input text corpus and is based on the explore and exploit paradigm: starting from a few seeds (or an existing dictionary) it effectively discovers new instances (explore) from the text corpus as well as predicts new potential instances which are not in the corpus, i.e. “unseen” , using the current dictionary entries (exploit). We evaluate our approach on five real-world dictionaries, achieving high accuracy with a rapid expansion rate.

In many Information Extraction tasks, dictionaries and lexica are powerful building blocks for sophisticated extractions. The success of the Semantic Web in the last 10 years has produced an unprecedented quantity of available structured data that can be leveraged to produce dictionaries on countless concepts in many domains. While being an invaluable resource, these automatically built dictionaries may contain "problematic" items, such as spurious words, which have been included by mistake, or ambiguous words, which appear with multiple different meanings in the target corpus and therefore necessitating an expensive disambiguation task. In this paper, we propose a simple and effective method to identify problematic terms in a given dictionary, which are ambiguous or spurious with respect to a given corpus, with the aim to facilitate subsequent Information Extraction tasks. We prove the effectiveness of the method with a systematic experiment on publicly available concept dictionaries, using a very large Web corpus as target, with an average precision in identifying a problem term above 85%.

Entity matching, or record linkage, is the task of identifying records that refer to the same entity.Naive entity matching techniques (i.e., brute-force pairwise comparisons) have quadratic complexity.A typical shortcut to the problem is to employ blocking techniques to reduce the number of comparisons, i.e. to partition the data in several blocks and only compare records within the same block.While classic blocking methods are designed for data from relational databases with clearly defined schemas, they are not applicable to data from Web tables, which are more prone to noise and do not come with an explicit schema.At the same time, Web tables are an interesting data source for many knowledge intensive tasks, which makes record linkage on Web Tables an important challenge.In this work, we propose an unsupervised approach to partition the data, that does not exploit any external knowledge, but only relies on heuristics to select the blocking attributes.We compare different partitioning methods: we use (i) clustering on bagof-words, (ii) binning via Locality-Sensitive Hashing and (iii) clustering using word embeddings.In particular, the clustering methods show good results on a standard dataset of Web Tables, and, when combined with word embeddings, are a robust solution which allows for computing the clusters in a dense, low-dimensional space.

Ontologies are dynamic artifacts that evolve both in structure and content. Keeping them up-to-date is a very expensive and critical operation for any application relying on semantic Web technologies. In this paper we focus on evolving the content of an ontology by extracting relevant instances of ontological concepts from text. We propose a novel technique which is (i) completely language independent, (ii) combines statistical methods with human-in-the-loop and (iii) exploits Linked Data as bootstrapping source. Our experiments on a publicly available medical corpus and on a Twitter dataset show that the proposed solution achieves comparable performances regardless of language, domain and style of text. Given that the method relies on a human-in-the-loop, our results can be safely fed directly back into Linked Data resources.

The Semantic Web Dog Food (SWDF) is the reference linked dataset of the Semantic Web community about papers, people, organisations, and events related to its academic conferences. In this paper we analyse the existing problems of generating, representing and maintaining Linked Data for the SWDF. With this work (i) we provide a refactored and cleaned SWDF dataset; (ii) we use a novel data model which improves the Semantic Web Conference Ontology, adopting best ontology design practices and (iii) we provide an open source workflow to support a healthy growth of the dataset beyond the Semantic Web conferences. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

This work explores the usage of Linked Data for Web scale Information Extraction and shows encouraging results on the task of Wrapper Induction. We propose a simple knowledge based method which is (i) highly flexible with respect to different domains and (ii) does not require any training material, but exploits Linked Data as background knowledge source to build essential learning resources. The major contribution of this work is a study of how Linked Data - an imprecise, redundant and large-scale knowledge resource - can be used to support Web scale Information Extraction in an effective and efficient way and identify the challenges involved. We show that, for domains that are covered, Linked Data serve as a powerful knowledge resource for Information Extraction. Experiments on a publicly available dataset demonstrate that, under certain conditions, this simple unsupervised approach can achieve competitive results against some complex state of the art that always depends on training data.

Abstract Measuring lexical semantic relatedness is an important task in Natural Language Processing (NLP). It is often a prerequisite to many complex NLP tasks. Despite an extensive amount of work dedicated to this area of research, there is a lack of an up-to-date survey in the field. This paper aims to address this issue with a study that is focused on four perspectives: (i) a comparative analysis of background information resources that are essential for measuring lexical semantic relatedness; (ii) a review of the literature with a focus on recent methods that are not covered in previous surveys; (iii) discussion of the studies in the biomedical domain where novel methods have been introduced but inadequately communicated across the domain boundaries; and (iv) an evaluation of lexical semantic relatedness methods and a discussion of useful lessons for the development and application of such methods. In addition, we discuss a number of issues in this field and suggest future research directions. It is believed that this work will be a valuable reference to researchers of lexical semantic relatedness and substantially support the research activities in this field.

An electronic performance support system (EPSS) introduces challenges on contextualized and personalized information delivery. Recommender systems aim at delivering and suggesting relevant information according to users preferences, thus EPSSs could take advantage of the recommendation algorithms that have the effect of guiding users in a large space of possible options. The JUMP project aims at integrating an EPSS with a hybrid recommender system. Collaborative and content-based filtering are the recommendation techniques most widely adopted to date. The main contribution of this paper is a content- collaborative hybrid recommender which computes similarities between users relying on their content- based profiles, in which user preferences are stored, instead of comparing their rating styles. A distinctive feature of our system is that a statistical model of the user interests is obtained by machine learning techniques integrated with linguistic knowledge contained in WordNet. This model, named "semantic user profile", is exploited by the hybrid recommender in the neighborhood formation process.