Search

Safapp is a tool that crawls and analyses data from social media (specially Twitter and blogs) in the context of radicalization detection. This app has been developed in the context of SAFFRON European project. This paper focuses on the description of the semantic module of Safapp which is dedicated to the analysis of textual content of social networks and blogs, and more specifically on a newly developed module: the event extractor. With this module we expect to go one step further in the development of useful tools to track and analyze online propaganda.

International audience; There has been increasing interest in the artificial intelligence community for influencer detection in recent years for its utility in singling out pertinent users within a large network of social media users. This could be useful, for example in commercial campaigns, to promote a product or a brand to a relevant target set of users. This task is performed either by analysing the graph-based representation of user interactions in a social network or by measuring the impact of the linguistic content of user messages in online discussions. We performed independent studies for each of these methods in the present paper with a hybridisation perspective. In the first study, we extract structural information to highlight influence among interaction networks. In the second, we identify linguistic features of influential behaviours. We then compute a score of user influence using centrality measures with the structural information for the former and a machine learning approach based on the relevant linguistic features for the latter.

Internet and Social media are widely used by terrorist organizations to spread their ideas and recruit foreign fighters. The aim of SAFFRON project is to build a system able to support early detection of foreign fighters' recruitment by terrorist groups in Europe. It consists in studying recruitment communication strategies on social media (e.g. narrations, argumentative tropes and myths used), and their evolution in time, as well as in identifying needs, values, cultural and social contexts of the target groups (young foreign fighters). In this paper, we will describe Safapp, the application developed to support semantic analysis of social network. We focus on how SAFFRON makes use of natural language processing and machine learning to categorize and analyse messages dealing with recruitment and radicalization on social networks.

SOMA is a smart social customer relationship management tool for companies aiming to monitor and automatically deal with customers’ complaints and interactions in social networks. Emotion detection is crucial in analyzing customers’ messages and to empathize with them. Part of the project consists in highlighting, in a social network corpus, how linguistic phenomena such as negation, intentional contexts, indirect speech, rhetorical devices, and pragmatic phenomena such as genre, personal style, domain, and context have a dramatic impact on the process of detecting opinions and emotions in social networks. We describe our approach to emotion detection based on a hybrid approach combining statistics and rules. A statistical approach is used to solve classic bag-of-words problems and a symbolic rule–based approach is used to increase detection precision. Finally, we propose a new operational evaluation method grounded more in real-world problem solving than in traditional gold standard annotation approaches.

Introduction The electronic health record (EHR) is a very important potential source of data for various areas, such as medical decision support tools, evidence-based medicine or epidemiological surveillance. Much of this data is available in text format. Methods of natural language processing can be used to perform data mining and facilitate interpretation. The purpose of this project was to develop a generic semantic solution for extracting and structuring medical data for epidemiological analyses or for medical decision-support. The solution was developed with the objective of making it as independent as possible from the field of medical application in order to allow any new user to write his or her own expert rules regardless of their area of medical expertise. Material and methods SYNODOS offers a modular architecture that makes a clear distinction between the linguistic rules and the medical expert rules. Different modules have been developed or adapted for this purpose: an interface between the multi-terminology server and semantic analyzer during the extraction phase, linguistic rules to extract temporal expressions, expert rules adapted to two areas of application (nosocomial infections, cancer), an interface between the engine and the linguistic knowledge base. Results Modular integrations were performed consecutively. The multi-terminology extractor and semantic analyzer were first interfaced during the extraction phase. Output of this data processing was then integrated into a knowledge base. A user interface to access documents and write business rules was developed. Expert rules for the detection of nosocomial infections and for the evaluation of colon cancer management have been developed. It was necessary to develop an additional module the need for which had not been identified during the drafting of the protocol. This module aims to structure the output of the data processing described above, according to the patient's care pathway. This module is based on the writing of medical expert rules. Evaluation indicators were obtained at different stages of the process (terminology extraction, semantic relations, data structuring, detection of events of interest). Discussion This project helped to highlight the value of combining different technologies (natural language processing, terminology, expert systems integration) to allow for the use of unstructured data in epidemiology. However, the need to develop an additional module of expert rules did not allow a complete and operational solution. Furthermore the multi-terminology extractor (ECMT V2) response time is too long (6 min per report). A change in technology was envisaged at the end of the project to reduce this time. Conclusions The originality of the SYNODOS project is the development of a single solution that integrates different technologies needed for the production of epidemiological indicators in the context of hospital activity. The project results confirm the interest but certain technological obstacles concerning the processing time need to be resolved in order to render the solution operational in a hospital environment.

This paper presents an unsupervised, all-words, word sense disambiguation system for English. The system associates a word with its meaning in a given context using an electronic dictionary as a tagged corpora in order to extract semantic disambiguation rules. The methodology attempts to avoid the data acquisition bottleneck observed in word sense disambiguation techniques. Semantic rules are used as input of a semantic application program encoding a linguistic strategy in order to select the best rule to apply. The semantic rule extraction process as well as the application program is described. The methodology is developed in a client/server architecture, which enables the treatment of large corpora. The evaluation of the system is then detailed and some possible extensions and perspectives are finally proposed.

Research on a number of developments in language technologies, targeted at improving patent processing procedures within patent offices and in subsequent patent database search systems, is described. Aspects of patent processing covered are (1) OCR correction, to assist the conversion of paper documents to electronic versions, and (2) text classification, to assist in the allocation of new patent applications to the correct technical experts. Aspects of patent searching covered are (3) terminology enrichment, linking keywords to IPC terms, (4) table and figure reference extraction, to enable the corresponding text to be located, and (5) multilingual information access, to help formulate queries in other languages for the search of documents in those languages, and then to provide contextual translation of key sections of the documents retrieved. The authors conclude that these types of technologies will support more sophisticated search activities, for example in patent information centres, to complement the low-cost or free offerings on the Internet.

1. Setting Up the French ExerciseTo make the evaluation exercise valuable and useful it is important to prepare theevaluation material according to a rigorous methodology. This includes havingclear criteria for choosing the words, being aware of the consequences on the eval-uation of the dictionary and corpus choice. Also, because sense disambiguation is adifficult task even for human beings, it is important to provide comparison figureswith human tagger agreement. In the following sections we present the adoptedmethodology together with the material used.1.1.

1. IntroductionThe French Semantic Dictionary Look-up (SDL) uses dictionary information aboutsubcategorization and collocates to perform Word Sense Disambiguation (WSD).The SDL is fully integrated in a multilingual comprehension system which uses theOxford Hachette French-English bilingual dictionary (OUP-H). Although the SDLworks on all words both for French and English, Romanseval results are relevantfor French verbs only because subcategorisation and collocate information is richerfor this part of speech in the OUP-H. The SDL uses dictionaries as semanticallytagged corpora of different languages, making the methodology reusable for anylanguage with existing on-line dictionaries.This paper first describes the system architecture as well as its components andresources. Second, it presents the work we did within Romanseval, namely sensemapping and results analysis.2. Semantic Dictionary Look-Up: Goal, Architecture and ComponentsThe SDL selects the most appropriate translation of a word appearing in a givencontext. It reorders dictionary entries making use of dictionary information. It isbuilt on top of Locolex

We have designed a system to support collaborative legal case reasoning and building. The design is based on our understanding of the corporate litigation domain acquired through analysis of the literature, interviews of various parties involved in corporate litigation processes, and studies of the commercial tools already available. In this paper we illustrate the designed system and in particular the interaction modes that it supports that we believe address a number of the requirements that emerged through our analysis. We also describe its main components and their integration, including a knowledge model that represents the domain, and a natural language processing component for extracting semantic information. A description of a prototype system is also provided.

Currently, extraction of data for epidemiological surveillance is severely limited by the lack of standardised structuring of medical records. The objective of this study is to describe a semi-automatic method for standardized coding of textual medical records for epidemiological use. In the context of the ALADIN-DTH research project, we are planning to develop a tool for detecting nosocomial infections. With that goal in mind we will ask physicians to manually code 2,000 hospital medical records using different appropriate medical terminologies (ICD10, SNOMED 3.5, ATC, CCAM, MeSH). A French automatic Multi-Terminology Health Concept Extractor (French acronym: ECMT) can offer a choice of labels and codes of different terminologies. This tool can be called on a distant server via an Internet connection thanks to an XML service. Among 3,450 medical expressions queried by users, the ECMT has proposed relevant codes for 70,5% of them (original formulation) and 87,7% of them after correction of the formulation by the annotator, this result ranging from 51,3% (bacteriological examinations) to 96,4% (symptoms/diagnoses). A multi-terminology health concept extractor is an interesting tool for standardized coding of textual data for epidemiological use.

Anonymization of personal data is a legal requirement for their use as part of a research project. In the context of developing a tool for detecting hospital-acquired infections, 2000 medical documents were needed for the research project ALADIN. To help annotators to anonymize this corpus of documents, a tool for the anonymization has been developed, relying on Natural Language Processing techniques. The recall, precision and F-score of the automatic phase of the anonymizer were respectively 79.7, 85.2 and 82.4%. The gold- standard used for the evaluation was the manual anonymization of the documents. The performance of the automatic anonymization can still be improved but the tool is already a considerable help in this process in terms of saving time and in terms of quality of anonymization (including the accuracy of labeling anonymized terms and computation of time duration).

We are interested in developing tools to support the activities of lawyers in corporate litigation. In current applications, information such as characters that have played a significant role in a case, events in which they have participated, people they have been in contact, etc., have to be manually identified. There is little in the way of support to help them identify the relevant information in the first place. In this paper, we describe an approach to semi-automatically extracting such information from the collection of documents the lawyers are searching. Our approach is based on Natural Language Processing techniques and it enables the use of entity related information corresponding to the relations among the key players of a case, extracted in the form of events.

Hospital Acquired Infections (HAI) is a real burden for doctors and risk surveillance experts. The impact on patients’ health and related healthcare cost is very significant and a major concern even for rich countries. Furthermore required data to evaluate the threat is generally not available to experts and that prevents from fast reaction. However, recent advances in Computational Intelligence Techniques such as Information Extraction, Risk Patterns Detection in documents and Decision Support Systems allow now to address this problem.

E-learning paves the way to a new type of course, more student centred, granulized, on demand, and highly interactive. Natural Language Processing (NLP) technologies associated with other multimedia technologies can help to address the major issues raised by this new type of courses: interaction, personalization and reliable information access. This paper presents Exills, a true e-learning solution which integrates natural language processing tools and virtual reality. Exills is unique in that unlike most of the language learning systems, it focuses on improving learners' performance rather than learners' competence.

In this paper we describe a method for performing word sense disambiguation (WSD). The method relies on unsupervised learning and exploits functional relations among words as produced by a shallow parser. By exploiting an error driven rule learning algorithm (Brill 1997), the system is able to produce rules for WSD, which can be optionally edited by humans in order to increase the performance of the system.

Most multi-word lexemes (MWLs) allow certain types of variation. This has to be taken into account for their description and their recognition in texts. We suggest to describe their syntactic restrictions and their idiosyncratic peculiarities with local grammar rules, which at the same time allow to express in a general way regularities valid for a whole class of MWLs. The local grammars can be written in a very convenient and compact way as regular expressions in the formalism IDAREX which uses a two-level morphology. IDAREX allows to define various types of variables, and to mix canonical and inflected word forms in the regular expressions.

Publisher Summary This chapter discusses the concept of dynamic relaxation. The fact that most parsers perform poorly on large corpora is widely acknowledged. To extend the scope of the grammar, the chapter presents the notions of dynamic relaxation and structural distance, based on the relaxed approach of the PLNLP broad coverage systems. Adopting a corpus-based paradigm implies confrontation with the gap among languages as theoretically and computationally described, and as attested by real-life usage. There is only partial overlapping between the structures actually observed in corpora and those usually described in grammar theories and dealt with by NLP systems. The discrepancies between theoretical or computational models and actual language usage can be seen as resulting from the distinction between the core and the periphery of language proposed within the generative paradigm. Following this approach, the grammar consists of a core and a periphery, where the core determines the general properties of the grammar, and the periphery accommodates the exceptional or marked properties.

To deal with long-distance dependencies, Applicative Universal Grammar (AUG) proposes a new type of categorial rules, called superposition rules. We compare the AUG rules with the alternative rules of Steedman's Combinatory Categorial Grammar (CCG) (Steedman, 1987, 1988, 1990; Szabolcsi, 1987; Ades and Steedman, 1982). In contrast to Steedman's rules, the AUG rules are free from inconsistencies in their semantic interpretation, free from spurious ambiguity. The superposition rules are based on the Theory of Type Superposition, established independently of the problem of long-distance dependencies and having a broad unifying power.

At the level of semantic relations, we are interested in finding the semantic links hidden in the syntax of a sentence. This involves, among other things, normalizing semantic structures across a wide range of paraphrases. The goal is achieved by taking the output of the preceding analysis components and modifying it with a “concept grammar,” written in PLNLP. The rules of this grammar are similar in form to the rules of preceding components; but they operate on different aspects of the common information structure, analyzing the relations between nodes in the sentence graph, and normalizing semantic structures and lexical relationships in a variety of syntactic domains, without losing access to the surface syntactic differences. This chapter shows how, starting from the argument structure output from PEGASUS, the concept grammar produces semantic graphs that preserve the broad-coverage, broad-domain characteristics of the entire system.