Search

Knowledge graphs are proving to be an increasingly important part of modern enterprises, and new applications of such enterprise knowledge graphs are still being found. In this paper, we report on the experience with the use of an automatic knowledge graph system called Saffron in the context of a large financial enterprise and show how this has found applications within this enterprise as part of the “Conversation Concepts Artificial Intelligence” tool. In particular, we analyse the use cases for knowledge graphs within this enterprise, and this led us to a new extension to the knowledge graph system. We present the results of these adaptations, including the introduction of a semi-supervised taxonomy extraction system, which includes analysts in-the-loop. Further, we extend the kinds of relations extracted by the system and show how the use of the BERTand ELMomodels can produce high-quality results. Thus, we show how this tool can help realize a smart enterprise and how requirements in the financial industry can be realised by state-of-the-art natural language processing technologies.

We discuss a domain-specific retrieval application for matching job seekers with open positions that uses a novel syntactic method of extracting skill-terms from the text of natural language job advertisements. Our new method is contrasted with two word embeddings methods, using word2vec. We define the notion of a skill headword, and present an algorithm that learns syntactic dependency patterns to recognize skill-terms. In all metrics, our syntactic method outperforms both word embeddings methods. Moreover, the word embeddings approaches were unable to model a meaningful distinction between skill-terms and non-skill-terms, while our syntactic approach was able to perform this successfully. We also show how these extracted skills can be used to automatically construct a semantic job-skills ontology, and facilitate a job-to-candidate matching system.

Purpose A hybrid approach is presented, which combines linguistic and statistical information to semi-automatically extract multiword term candidates from texts. Design/methodology/approach The method is designed to be domain and language independent, focusing on languages with rich morphology. Here, it is used for extracting multiword terms from texts in Serbian, belonging to the agricultural engineering domain, as a use case. Predefined syntactic structures were used for multiword terms. For each structure, a finite state transducer was developed, which recognizes text sequences having that structure and outputs the sequence in a normalized form, so that different inflectional forms of the same multiword term can be counted properly. Term candidates were further filtered by their frequencies and evaluated by two domain experts. Findings By using language resources, such as electronic dictionaries and grammars, 928 multiword terms were extracted out of 1,523 multiword terms that were recognized as candidates from a corpus having 42,260 different simple word forms; 870 of these were new, not already contained in the existing electronic dictionary of compounds for Serbian, and they were used to enrich the dictionary. Originality/value The paper presents methodology that can significantly contribute to the development of terminology lexicons in different areas. In this particular use case, some important agricultural engineering concepts were extracted from the text, but this approach could be used for other domains and languages as well.

In the present era of Big Data the demand for developing efficient information processing techniques for different applications is expanding steadily. One such possible application is automatic creation of ontology. Such an ontology is often found to be helpful for answering queries for the underlying domain. The present work proposes a scheme for designing an ontology for agriculture domain. The proposed scheme works in two steps. In the first step it uses domain-dependent regular expressions and natural language processing techniques for automatic extraction of vocabulary pertaining to agriculture domain. In the second step semantic relationships between the extracted terms and phrases are identified. A rule-based reasoning algorithm RelExOnt has been proposed for the said task. Human evaluation of the term extraction output yields precision and recall of 75.7% and 60%, respectively. The relation extraction algorithm, RelExOnt performs well with an average precision of 86.89%.

Very short texts, such as tweets and invoices, present challenges in classification. Although term occurrences are strong indicators of content, in very short texts, the sparsity of these texts makes it difficult to capture important semantic relationships. A solution calls for a method that not only considers term occurrence, but also handles sparseness well. In this work, we introduce such an approach, the Term Based Semantic Clusters (TBSeC) that employs terms to create distinctive semantic concept clusters. These clusters are ranked using a semantic similarity function which in turn defines a semantic feature space that can be used for text classification. Our method is evaluated in an invoice classification task. Compared to well-known content representation methods the proposed method performs competitively.

Traditional approaches to automatic term extraction do not rely on machine learning (ML) and select the top n ranked candidate terms or candidate terms above a certain predefined cut-off point, based on a limited number of linguistic and statistical clues. However, supervised ML approaches are gaining interest. Relatively little is known about the impact of these supervised methodologies; evaluations are often limited to precision, and sometimes recall and f1-scores, without information about the nature of the extracted candidate terms. Therefore, the current paper presents a detailed and elaborate analysis and comparison of a traditional, state-of-the-art system (TermoStat) and a new, supervised ML approach (HAMLET), using the results obtained for the same, manually annotated, Dutch corpus about dressage. ispartof: pages:1012-1021 ispartof: Proceedings of Recent Advances in Natural Language Processing (RANLP 2019) : natural language processing in a deep learining world pages:1012-1021 ispartof: Recent Advances in Natural Language Processing (RANLP 2019) location:Varna, Bulgaria date:2 Sep - 4 Sep 2019 status: published

Tracking developments in the highly dynamic data-technology landscape are vital to keeping up with novel technologies and tools, in the various areas of Artificial Intelligence (AI). However, It is difficult to keep track of all the relevant technology keywords. In this paper, we propose a novel system that addresses this problem. This tool is used to automatically detect the existence of new technologies and tools in text, and extract terms used to describe these new technologies. The extracted new terms can be logged as new AI technologies as they are found on-the-fly in the web. It can be subsequently classified into the relevant semantic labels and AI domains. Our proposed tool is based on a two-stage cascading model -- the first stage classifies if the sentence contains a technology term or not; and the second stage identifies the technology keyword in the sentence. We obtain a competitive accuracy for both tasks of sentence classification and text identification., Published in 11th International Conference on Computer and Automation Engineering (ICCAE 2019)

This study intends to account for the process involved in the construction of the conceptual meaning of verbs (#EVENTS) directly related to legal aspects of terrorism and organized crime based on the evidence provided by the Globalcrimeterm Corpus and the consistent application of specific criteria for term extraction. The selected 49 concepts have eventually been integrated in the Core Ontology of FunGramKB (Functional Grammar Knowledge Base), a knowledge base which is founded on the principles of deep semantics and is also aimed at the computational development of the Lexical Constructional Model (www.fungramkb.com). To achieve this purpose, key phases of the COHERENT methodology (Periñán Pascual & Mairal Usón 2011) are followed, particularly those which involve the modelling, subsumption and hierarchisation of the aforementioned verbal concepts. The final outcome of this research shows that most of the apparently specialised conceptual units should eventually be included in the Core Ontology instead of the specific Globalcrimeterm Subontology, due to the fact that the semantic content of their corresponding lexical units can be found in widely used learner`s dictionaries and, consequently, this conceptual information is not only shared by the experts in the field but also by the layperson and the average speaker of the language.

Constructing an ontology for terminological purposes involves several steps: among them, to acquire the knowledge necessary to create the ontology, to conceptualize the domain and to implement the ontology itself. This paper focuses on the first activity and proposes a protocol to work from a specialized corpus, extract terms, detect linguistic equivalents and extract conceptual relations, with a view to creating an ontology, which in turn can be the basis for a multilingual ontoterminological dictionary.

Linguistic phenomena associated with the analysis of document content and employed for the purpose of organization and retrieval are well-visited objects of study in the field of library and information science. Language often acts as a gatekeeper, admitting or excluding people from gaining access to knowledge. As such, the terms used in the scientific and technical language of research need to be kept up and their behavior within the domain examined. Documental content analysis of scientific texts provides knowledge of specialized lexicons and their specific applications, while differentiating them from common use in order to establish indexing languages. Thus, as proposed herein, the application of lexicographic techniques to documental content analysis of non-specialized language yields the components needed to describe and extract lexical units of the specialized language.

Existing term extraction systems have predominantly targeted large and well-written document collections, which provide reliable statistical and linguistic evidence to support term extraction. In this article, we address the term extraction challenges posed by sparse, ungrammatical texts with domain-specific contents, such as customer complaint emails and engineers' repair notes. To this aim, we present ExtTerm, a novel term extraction system. Specifically, as our core innovations, we accurately detect rare (low frequency) terms, overcoming the issue of data sparsity. These rare terms may denote critical events, but they are often missed by extant TE systems. ExtTerm also precisely detects multi-word terms of arbitrarily lengths, e.g. with more than 2 words. This is achieved by exploiting fundamental theoretical notions underlying term formation, and by developing a technique to compute the collocation strength between any number of words. Thus, we address the limitation of existing TE systems, which are primarily designed to identify terms with 2 words. Furthermore, we show that open-domain (general) resources, such as Wikipedia, can be exploited to support domain-specific term extraction. Thus, they can be used to compensate for the unavailability of domain-specific knowledge resources. Our experimental evaluations reveal that ExtTerm outperforms a state-of-the-art baseline in extracting terms from a domain-specific, sparse and ungrammatical real-life text collection. (c) 2012 Elsevier Ltd. All rights reserved.

[Abstract] Entity Linking (EL) consists in linking name mentions in a given text with their referring entities in external knowledge bases such as DBpedia/Wikipedia. In this paper, we propose an EL approach whose main contribution is to make use of a knowledge base built by means of distributional similarity. More precisely, Wikipedia is transformed into a manageable database structured with similarity relations between entities. Our EL method is focused on a specific task, namely semantic annotation of documents by extracting those relevant terms that are linked to nodes in DBpedia/Wikipedia. The method is currently working for four languages. The Portuguese and English versions have been evaluated and compared against other EL systems, showing competitive range, close to the best systems Ministerio de Economía y Competitividad; FFI2014-51978-C2-1-R

We evaluate five term scoring methods for automatic term extraction on four different types of text collections: personal document collections, news articles, scientific articles and medical discharge summaries. Each collection has its own use case: author profiling, boolean query term suggestion, personalized query suggestion and patient query expansion. The methods for term scoring that have been proposed in the literature were designed with a specific goal in mind. However, it is as yet unclear how these methods perform on collections with characteristics different than what they were designed for, and which method is the most suitable for a given (new) collection. In a series of experiments, we evaluate, compare and analyse the output of six term scoring methods for the collections at hand. We found that the most important factors in the success of a term scoring method are the size of the collection and the importance of multi-word terms in the domain. Larger collections lead to better terms; all methods are hindered by small collection sizes (below 1000 words). The most flexible method for the extraction of single-word and multi-word terms is pointwise Kullback---Leibler divergence for informativeness and phraseness. Overall, we have shown that extracting relevant terms using unsupervised term scoring methods is possible in diverse use cases, and that the methods are applicable in more contexts than their original design purpose.

The extraction of multi-word relevant expressions has been an increasingly hot topic in the last few years. Relevant expressions are applicable in diverse areas such as Information Retrieval, document clustering, or classification and indexing of documents. However, relevant single-words, which represent much of the knowledge in texts, have been a relatively dormant field. In this paper we present a statistical language-independent approach to extract concepts formed by relevant single and multi-word units. By achieving promising precision/recall values, it can be an alternative both to language dependent approaches and to extractors that deal exclusively with multi-words.

One of the many implications of the process of language democratization which started post-1994 in South Africa is the empowerment of the previously marginalized South African Bantu languages to become languages of higher functions, i.e. languages of learning and teaching, and also of scientific discourse. This in turn implies the development, consolidation and especially standardization of terminology for each of these languages, and the compilation of LSP dictionaries. This article describes the terminological processing of a technical source text prior to translation, which formed part of the compilation of a Quadrilingual Explanatory Dictionary of Chemistry . It reports on the model of terminology management that was utilized and explores strategies for the internal standardization of terms in the absence of readily available, standardized chemistry terminology.

The UK Education Evidence Portal (eep) provides a single, searchable, point of access to the contents of the websites of 33 organizations relating to education, with the aim of revolutionizing work practices for the education community. Use of the portal alleviates the need to spend time searching multiple resources to find relevant information. However, the combined content of the websites of interest is still very large (over 500 000 documents and growing). This means that searches using the portal can produce very large numbers of hits. As users often have limited time, they would benefit from enhanced methods of performing searches and viewing results, allowing them to drill down to information of interest more efficiently, without having to sift through potentially long lists of irrelevant documents. The Joint Information Systems Committee (JISC)-funded ASSIST project has produced a prototype web interface to demonstrate the applicability of integrating a number of text-mining tools and methods into the eep, to facilitate an enhanced searching, browsing and document-viewing experience. New features include automatic classification of documents according to a taxonomy, automatic clustering of search results according to similar document content, and automatic identification and highlighting of key terms within documents. This journal is © 2010 The Royal Society.

The need for domain ontologies motivates the research on structured information extraction from texts. A foundational part of this process is the identification of domain relevant compound terms. This paper presents an evaluation of compound terms extraction from a corpus of the domain of Pediatrics. Bigrams and trigrams were automatically extracted from a corpus composed by 283 texts from a Portuguese journal, Jornal de Pediatria, using three different extraction methods. Considering that these methods generate an elevated number of candidates, we analyzed the quality of the resulting terms according to different methods and cut-off points. The evaluation is reported by metrics such as precision, recall and f-measure, which are computed on the basis of a hand-made reference list of domain relevant compounds.

Term extraction is an important task that automatically extracts relative terms from the texts in a given domain. A significant number of web applications need to model information for specific topics. In particular, we have explored a Taiwan government website that maintains the Laws &amp, Regulations Database of the Republic of China (R.O.C) to provide the current Chinese law text to the public. However, the main issue is that there is no efficient structured method to handle such a large number of law texts. Therefore, in this paper, we propose a novel approach to extract legal as well as domain-relative terms, and then build a law ontology. We used the current Chinese law text from the Laws &amp, Regulations Database as the data source. We then utilized natural language processing tools and data mining techniques to extract legal keywords and their definitions automatically. Subsequently, we constructed a Taiwan law ontology with the legal keywords and relative definitions. We have extracted 1114 legal keywords with definitions. With the characteristics of an ontology, users can view one keyword with its information and the associated keywords. Furthermore, we provide a service, which includes both the graphical and text interfaces to users on the web, such that a user can readily access the legal information on the Internet.

Among the challenges in developing terminology systems is providing complete content coverage of specialized subject fields. This paper reports on a term extraction tool designed for the development and expansion of terminology systems concerned with functioning, disability, and health. Content relevant to this domain is the emphasis of the foci and targets of many nursing terminologies. We extend previously published term extraction algorithms by applying two filters. The first filter is based on the raw frequency of the content words in the lexical string under consideration. The second filter applies the notion of a complete syntactic node to discover relevant noun or verb phrases. While we report on a limited corpus (30,607 words comprising 4103 terms from 60 dismissal note summaries), the recall, precision, and F-measures we observed are encouraging and suggest continued development and testing of the tool is merited.

International audience; Traiter des questions factuelles portant sur n’importe quel sujet dont la réponse appartient à un très large corpus entraîne l’application de processus de Traitement Automatique des Langues (TAL) afin de localiser la réponse dans les documents sélectionnés par un moteur de recherche. Le système QALC, présenté dans cet article, a participé à la tâche Question Answering lors des évaluations TREC8 et TREC9. QALC exploite les résultats d’une analyse des documents où les termes de la question traitée, ainsi que leurs variantes, sont étiquetés et pondérés. Ces éléments permettent de pondérer les documents afin de n’en sélectionner qu’une partie pour la suite des traitements. Par ailleurs, ils permettent aussi d’améliorer la sélection des phrases candidates lorsqu’elles sont comparées à une question afin de décider si elles sont susceptibles de contenir la réponse. Cette comparaison exploite également les résultats d’une analyse des questions indiquant le type de la réponse attendue, qui est rapproché du marquage, dans les phrases, des types d’entités nommées reconnues par le système.

Supervised text classifiers need to learn from many labeled examples to achieve a high accuracy. However, in a real context, sufficient labeled examples are not always available because human labeling is enormously time-consuming. For this reason, there has been recent interest in methods that are capable of obtaining a high accuracy when the size of the training set is small. In this paper we introduce a new single label text classification method that performs better than baseline methods when the number of labeled examples is small. Differently from most of the existing methods that usually make use of a vector of features composed of weighted words, the proposed approach uses a structured vector of features, composed of weighted pairs of words. The proposed vector of features is automatically learned, given a set of documents, using a global method for term extraction based on the Latent Dirichlet Allocation implemented as the Probabilistic Topic Model. Experiments performed using a small percentage of the original training set (about 1%) confirmed our theories. © 2013 Elsevier Ltd. All rights reserved.

International audience; In this paper, we study terminology extraction in all its states. We present two terminology extraction tools to compare a knowledge-poor vs. a knowledge-rich approach. Both tools deal with single and multi-word terms and are designed to handle multilingualism. Today they can process six languages, a new language can be added quite rapidly. We present an evaluation run on the six languages and two different domains using crawled comparable corpora, as well as hand-crafted reference term lists. We discuss the three main results for the terminology extraction issue. The first two scenarios concern the knowledge-rich framework. First we compare the performances of the ranking and compare the use of the specificity score vs. the the number of occurrences for all the languages. Secondly we examine the relevancy of the term variant identification to increase the precision ranking whatever the language is. The third evaluation scenario compares the two tools and shows that a probabilistic term extraction approach developed with a minimal effort achieves satisfactory results compared to a rule-based method.

It is well known that supervised text classification methods need to learn from many labeled examples to achieve a high accuracy. However, in a real context, sufficient labeled examples are not always available. In this paper we demonstrate that a way to obtain a high accuracy, when the number of labeled examples is low, is to consider structured features instead of list of weighted words as observed features. The proposed vector of features considers a hierarchical structure, named a mixed Graph of Terms, composed of a directed and an undirected sub-graph of words, that can be automatically constructed from a set of documents through the probabilistic Topic Model. © Springer-Verlag Berlin Heidelberg 2013.

Various information extraction (IE) systems for corporate usage exist. However, none of them target the product development and/or customer service domain, despite significant application potentials and benefits. This domain also poses new scientific challenges, such as the lack of external knowledge resources, and irregularities like ungrammatical constructs in textual data, which compromise successful information extraction. To address these issues, we describe the development of Textractor; an application for accurately extracting relevant concepts from irregular textual narratives in datasets of product development and/or customer service organizations. The extracted information can subsequently be fed to a host of business intelligence activities. We present novel algorithms, combining both statistical and linguistic approaches, for the accurate discovery of relevant domain concepts from highly irregular/ungrammatical texts. Evaluations on real-life corporate data revealed that Textractor extracts domain concepts, realized as single or multi-word terms in ungrammatical texts, with high precision.

International audience; The PERTOMed project is a pluri-disciplinary research initiative undertaken by several institutions in France. Applications considered within the part of the project described in this article concern pharmacovigilance and adverse drug reactions. We had multiple objectives: to create a specialized Russian Internet corpus; to test new tools and methods for term extraction from comparable multilingual texts and to build terminological resources including Russian. A trilingual Russian-French-English lexicon resulting form this work is freely available from the PERTOMed server.

Summarization: AMTEx is a medical document indexing method, specifically designed for the automatic indexing of documents in large medical collections, such as MEDLINE, the premier bibliographic database of the US National Library of Medicine (NLM). AMTEx combines MeSH, the terminological thesaurus resource of NLM, with a well-established method for extraction of terminology, the C/NC-value method. The performance evaluation of two AMTEx configurations is measured against the current state-of-the-art, the MetaMap Transfer (MMTx) method in four experiments, using two types of corpora: a subset of MEDLINE (PMC) full document corpus and a subset of MEDLINE (OHSUMED) abstracts, for each of the indexing and retrieval tasks, respectively. The experimental results demonstrate that AMTEx performs better in indexing in 20–50% of the processing time compared to MMTx, while for the retrieval task, AMTEx performs better in the full text (PMC) corpus. Presented on: Data and Knowledge Engineering

This paper presents a novel methodology for topic ontology learning from text documents. The proposed methodology, named OntoTermExtraction (Term Extraction for Ontology learning), is based on OntoGen, a semi-automated tool for topic ontology construction, upgraded by using an advanced terminology extraction tool in an iterative, semi-automated ontology construction process. This process consists of (a) document clustering to find the nodes in the topic ontology, (b) term extraction from document clusters, (c) populating the term vocabulary and keyword extraction, and (d) choosing the concept names by comparing the best-ranked terms with the extracted keywords. The approach was successfully used for generating the ontology of topics in Inductive Logic Programming, learned semi-automatically from papers indexed in the ILPnet2 publications database.

This paper presents an original natural language processing (NLP) approach for building of back-of-the-book indexes. Our indexing system, IndDoc, exploits some terminological tools and automatically builds an index draft of the analysis of the document text. The indexer then has to validate that index draft through a dedicated interface. This approach has been tested on several documents with promising results. Relying on our experience in developing and testing the IndDoc indexing system, we aim at assessing the contribution of terminological analysis as well as the level of maturity that computational terminology has reached in the indexing perspective.

This paper presents how we utilise natural language processing techniques in order to “automagically” classify information stored in a CRIS, and aggregate the information in a researchers portfolio into a “fingerprint” describing a researchers research interest. Our approach exploits the fact that entities in a CRIS typically include some kind of text – most notable example being publication abstracts. We explain how the approach can result in automatic detailed classification of information, and argue how we can take advantage of such information in order to facilitate networking. Finally, we describe how we have realised the solution within our CRIS system.

Unstructured metadata fields such as 'description' offer tremendous value for users to understand cultural heritage objects. However, this type of narrative information is of little direct use within a machine-readable context due to its unstructured nature. This article explores the possibilities and limitations of named-entity recognition (NER) and term extraction (TE) to mine such unstructured metadata for meaningful concepts. These concepts can be used to leverage otherwise limited searching and browsing operations, but they can also play an important role to foster Digital Humanities research. To catalyze experimentation with NER and TE, the article proposes an evaluation of the performance of three third-party entity extraction services through a comprehensive case study, based on the descriptive fields of the Smithsonian Cooper-Hewitt National Design Museum in New York. To cover both NER and TE, we first offer a quantitative analysis of named entities retrieved by the services in terms of precision and recall compared with a manually annotated gold-standard corpus, and then complement this approach with a more qualitative assessment of relevant terms extracted. Based on the outcomes of this double analysis, the conclusions present the added value of entity extraction services, but also indicate the dangers of uncritically using NER and/or TE, and by extension Linked Data principles, within the Digital Humanities. All metadata and tools used within the article are freely available, making it possible for researchers and practitioners to repeat the methodology. By doing so, the article offers a significant contribution towards understanding the value of entity recognition and disambiguation for the Digital Humanities.