DOCSLIB.ORG

Vector Space Model

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Vector Space Model Vector space model Vector space model or term vector model is an alge- braic model for representing text documents (and any ob- jects, in general) as vectors of identifiers, such as, for ex- ample, index terms. It is used in information filtering, information retrieval, indexing and relevancy rankings. Its first use was in the SMART Information Retrieval Sys- tem. 1 Definitions Documents and queries are represented as vectors. dj = (w1;j ; w2;j; : : : ; wt;j) q = (w1;q; w2;q; : : : ; wn;q) Each dimension corresponds to a separate term. If a term occurs in the document, its value in the vector is non- zero. Several different ways of computing these values, (q in the figure) vectors, kd2k is the norm of vector d2, also known as (term) weights, have been developed. One and kqk is the norm of vector q. The norm of a vector is of the best known schemes is tf-idf weighting (see the calculated as such: example below). v The definition of term depends on the application. Typi- u cally terms are single words, keywords, or longer phrases. uXn k k t 2 If words are chosen to be the terms, the dimensionality of q = qi the vector is the number of words in the vocabulary (the i=1 number of distinct words occurring in the corpus). As all vectors under consideration by this model are ele- Vector operations can be used to compare documents mentwise nonnegative, a cosine value of zero means that with queries. the query and document vector are orthogonal and have no match (i.e. the query term does not exist in the docu- ment being considered). See cosine similarity for further 2 Applications information. Relevance rankings of documents in a keyword search can be calculated, using the assumptions of document 3 Example: tf-idf weights similarities theory, by comparing the deviation of an- gles between each document vector and the original query In the classic vector space model proposed by Salton, vector where the query is represented as the same kind of Wong and Yang [1] the term-specific weights in the docu- vector as the documents. ment vectors are products of local and global parameters. In practice, it is easier to calculate the cosine of the angle The model is known as term frequency-inverse document between the vectors, instead of the angle itself: frequency model. The weight vector for document d is T vd = [w1;d; w2;d; : : : ; wN;d] , where d2 · q cos θ = k k k k jDj d2 q w = tf · log t;d t;d jfd0 2 D j t 2 d0gj Where d2 · q is the intersection (i.e. the dot product) of the document (d2 in the figure to the right) and the query and 1 2 7 SOFTWARE THAT IMPLEMENTS THE VECTOR SPACE MODEL • tft;d is term frequency of term t in document d (a 1. Long documents are poorly represented because local parameter) they have poor similarity values (a small scalar prod- uct and a large dimensionality) • jDj log jfd02D j t2d0gj is inverse document frequency (a global parameter). jDj is the total number of doc- 2. Search keywords must precisely match document uments in the document set; jfd0 2 D j t 2 d0gj is terms; word substrings might result in a "false posi- the number of documents containing the term t. tive match” 3. Semantic sensitivity; documents with similar con- Using the cosine the similarity between document dj and text but different term vocabulary won't be associ- query q can be calculated as: ated, resulting in a "false negative match”. P 4. The order in which the terms appear in the document · N w w dj q q i=1 qi;j i;q is lost in the vector space representation. sim(dj; q) = = P P kdjk kqk N 2 N 2 i=1 wi;j i=1 wi;q 5. Theoretically assumed terms are statistically inde- pendent. 4 Advantages 6. Weighting is intuitive but not very formal. The vector space model has the following advantages over Many of these difficulties can, however, be overcome the Standard Boolean model: by the integration of various tools, including mathemati- cal techniques such as singular value decomposition and 1. Simple model based on linear algebra lexical databases such as WordNet. 2. Term weights not binary 3. Allows computing a continuous degree of similarity 6 Models based on and extending between queries and documents the vector space model 4. Allows ranking documents according to their possi- Models based on and extending the vector space model ble relevance include: 5. Allows partial matching • Generalized vector space model Most of these advantages are a consequence of the dif- • ference in the density of the document collection repre- Latent semantic analysis sentation between Boolean and tf-idf approaches. When • Term Discrimination using Boolean weights, any document lies in a vertex in a n-dimensional hypercube. Therefore, the possible docu- • n Rocchio Classification ment representations are 2pand the maximum Euclidean distance between pairs is n . As documents are added • Random Indexing to the document collection, the region defined by the hypercube’s vertexes become more populated and hence denser. Unlike Boolean, when a document is added using tf-idf weights, the idfs of the terms in the new document 7 Software that implements the decrease while that of the remaining terms increase. In vector space model average, as documents are added, the region where docu- ments lie expands regulating the density of the entire col- lection representation. This behavior models the original The following software packages may be of interest to motivation of Salton and his colleagues that a document those wishing to experiment with vector models and im- collection represented in a low density region could yield plement search services based upon them. better retrieval results. 7.1 Free open source software 5 Limitations • Apache Lucene. Apache Lucene is a high- performance, full-featured text search engine library The vector space model has the following limitations: written entirely in Java. 3 • Gensim is a Python+NumPy framework for Vector Space modelling. It contains incremental (memory- efficient) algorithms for Tf–idf, Latent Semantic In- dexing, Random Projections and Latent Dirichlet Allocation. • Weka. Weka is a popular data mining package for Java including WordVectors and Bag Of Words models. 8 Further reading • G. Salton, A. Wong, and C. S. Yang (1975), "A Vec- tor Space Model for Automatic Indexing,” Commu- nications of the ACM, vol. 18, nr. 11, pages 613– 620. (Article in which a vector space model was pre- sented) • David Dubin (2004), The Most Influential Paper Gerard Salton Never Wrote (Explains the history of the Vector Space Model and the non-existence of a frequently cited publication) • Description of the vector space model • Description of the classic vector space model by Dr E. Garcia • Relationship of vector space search to the “k- Nearest Neighbor” search 9 See also • Bag-of-words model • Compound term processing • Conceptual space • Eigenvalues and eigenvectors • Inverted index • Nearest neighbor search • Sparse distributed memory • w-shingling 10 References [1] G. Salton , A. Wong , C. S. Yang, A vector space model for automatic indexing, Communications of the ACM, v.18 n.11, p.613-620, Nov. 1975 4 11 TEXT AND IMAGE SOURCES, CONTRIBUTORS, AND LICENSES 11 Text and image sources, contributors, and licenses 11.1 Text • Vector space model Source: https://en.wikipedia.org/wiki/Vector_space_model?oldid=744792705 Contributors: Michael Hardy, Kku, Dcljr, Stan Shebs, Jitse Niesen, Gdm, Beland, Thorwald, Rama, Mykhal, ESkog, Aaronbrick, .:Ajvol:., Jonsafari, Gary, Dominik Kuropka~enwiki, Bjh~enwiki, Ruud Koot, GregorB, Qwertyus, LanguageMan, Rjwilmsi, Gmelli, YurikBot, Conscious, Fmccown, Mike Dillon, SmackBot, MalafayaBot, Morecore~enwiki, JohnWhitlock, Stiang, Hankat, Padvi~enwiki, Thijs!bot, Oliver202, Remaire, AnAj, SamatJain, Jonemerson, Destynova, Ezani, Unkx80, Cometstyles, Dominich01, VolkovBot, Amroamroamro, Philip Trueman, Informa- tionSpace, Synthebot, Luc.denys, Disooqi, Maxalbanese, Dspattison, UKoch, PixelBot, Sir Tobek, Dwiddows, XLinkBot, Addbot, Favo- nian, Vuongvina, LatentDrK, Hyju, Suffusion of Yellow, TigerHokieFan, Riclas, Boraas, ZéroBot, Donner60, Tbear1234, PenelopeKit, Justincheng12345-bot, Lxcythian, Biogeographist, SergioJimenez, Alenrooni, Vítor and Anonymous: 54 11.2 Images • File:Vector_space_model.jpg Source: https://upload.wikimedia.org/wikipedia/commons/f/ff/Vector_space_model.jpg License: CC BY 3.0 Contributors: Own work Original artist: Riclas 11.3 Content license • Creative Commons Attribution-Share Alike 3.0.
Load more

Recommended publications
  • Document and Topic Models: Plsa
    10/4/2018 Document and Topic Models: pLSA and LDA Andrew Levandoski and Jonathan Lobo CS 3750 Advanced Topics in Machine Learning 2 October 2018 Outline • Topic Models • pLSA • LSA • Model • Fitting via EM • pHITS: link analysis • LDA • Dirichlet distribution • Generative process • Model • Geometric Interpretation • Inference 2 1 10/4/2018 Topic Models: Visual Representation Topic proportions and Topics Documents assignments 3 Topic Models: Importance • For a given corpus, we learn two things: 1. Topic: from full vocabulary set, we learn important subsets 2. Topic proportion: we learn what each document is about • This can be viewed as a form of dimensionality reduction • From large vocabulary set, extract basis vectors (topics) • Represent document in topic space (topic proportions) 푁 퐾 • Dimensionality is reduced from 푤푖 ∈ ℤ푉 to 휃 ∈ ℝ • Topic proportion is useful for several applications including document classification, discovery of semantic structures, sentiment analysis, object localization in images, etc. 4 2 10/4/2018 Topic Models: Terminology • Document Model • Word: element in a vocabulary set • Document: collection of words • Corpus: collection of documents • Topic Model • Topic: collection of words (subset of vocabulary) • Document is represented by (latent) mixture of topics • 푝 푤 푑 = 푝 푤 푧 푝(푧|푑) (푧 : topic) • Note: document is a collection of words (not a sequence) • ‘Bag of words’ assumption • In probability, we call this the exchangeability assumption • 푝 푤1, … , 푤푁 = 푝(푤휎 1 , … , 푤휎 푁 ) (휎: permutation) 5 Topic Models: Terminology (cont’d) • Represent each document as a vector space • A word is an item from a vocabulary indexed by {1, … , 푉}. We represent words using unit‐basis vectors.
    [Show full text]
  • Redalyc.Latent Dirichlet Allocation Complement in the Vector Space Model for Multi-Label Text Classification
    International Journal of Combinatorial Optimization Problems and Informatics E-ISSN: 2007-1558 [email protected] International Journal of Combinatorial Optimization Problems and Informatics México Carrera-Trejo, Víctor; Sidorov, Grigori; Miranda-Jiménez, Sabino; Moreno Ibarra, Marco; Cadena Martínez, Rodrigo Latent Dirichlet Allocation complement in the vector space model for Multi-Label Text Classification International Journal of Combinatorial Optimization Problems and Informatics, vol. 6, núm. 1, enero-abril, 2015, pp. 7-19 International Journal of Combinatorial Optimization Problems and Informatics Morelos, México Available in: http://www.redalyc.org/articulo.oa?id=265239212002 How to cite Complete issue Scientific Information System More information about this article Network of Scientific Journals from Latin America, the Caribbean, Spain and Portugal Journal's homepage in redalyc.org Non-profit academic project, developed under the open access initiative © International Journal of Combinatorial Optimization Problems and Informatics, Vol. 6, No. 1, Jan-April 2015, pp. 7-19. ISSN: 2007-1558. Latent Dirichlet Allocation complement in the vector space model for Multi-Label Text Classification Víctor Carrera-Trejo1, Grigori Sidorov1, Sabino Miranda-Jiménez2, Marco Moreno Ibarra1 and Rodrigo Cadena Martínez3 Centro de Investigación en Computación1, Instituto Politécnico Nacional, México DF, México Centro de Investigación e Innovación en Tecnologías de la Información y Comunicación (INFOTEC)2, Ags., México Universidad Tecnológica de México3 – UNITEC MÉXICO [email protected], {sidorov,marcomoreno}@cic.ipn.mx, [email protected] [email protected] Abstract. In text classification task one of the main problems is to choose which features give the best results. Various features can be used like words, n-grams, syntactic n-grams of various types (POS tags, dependency relations, mixed, etc.), or a combinations of these features can be considered.
    [Show full text]
  • Evaluating Vector-Space Models of Word Representation, Or, the Unreasonable Effectiveness of Counting Words Near Other Words
    Evaluating Vector-Space Models of Word Representation, or, The Unreasonable Effectiveness of Counting Words Near Other Words Aida Nematzadeh, Stephan C. Meylan, and Thomas L. Griffiths University of California, Berkeley fnematzadeh, smeylan, tom griffi[email protected] Abstract angle between word vectors (e.g., Mikolov et al., 2013b; Pen- nington et al., 2014). Vector-space models of semantics represent words as continuously-valued vectors and measure similarity based on In this paper, we examine whether these constraints im- the distance or angle between those vectors. Such representa- ply that Word2Vec and GloVe representations suffer from the tions have become increasingly popular due to the recent de- same difficulty as previous vector-space models in capturing velopment of methods that allow them to be efficiently esti- mated from very large amounts of data. However, the idea human similarity judgments. To this end, we evaluate these of relating similarity to distance in a spatial representation representations on a set of tasks adopted from Griffiths et al. has been criticized by cognitive scientists, as human similar- (2007) in which the authors showed that the representations ity judgments have many properties that are inconsistent with the geometric constraints that a distance metric must obey. We learned by another well-known vector-space model, Latent show that two popular vector-space models, Word2Vec and Semantic Analysis (Landauer and Dumais, 1997), were in- GloVe, are unable to capture certain critical aspects of human consistent with patterns of semantic similarity demonstrated word association data as a consequence of these constraints. However, a probabilistic topic model estimated from a rela- in human word association data.
    [Show full text]
  • Gensim Is Robust in Nature and Has Been in Use in Various Systems by Various People As Well As Organisations for Over 4 Years
    Gensim i Gensim About the Tutorial Gensim = “Generate Similar” is a popular open source natural language processing library used for unsupervised topic modeling. It uses top academic models and modern statistical machine learning to perform various complex tasks such as Building document or word vectors, Corpora, performing topic identification, performing document comparison (retrieving semantically similar documents), analysing plain-text documents for semantic structure. Audience This tutorial will be useful for graduates, post-graduates, and research students who either have an interest in Natural Language Processing (NLP), Topic Modeling or have these subjects as a part of their curriculum. The reader can be a beginner or an advanced learner. Prerequisites The reader must have basic knowledge about NLP and should also be aware of Python programming concepts. Copyright & Disclaimer Copyright 2020 by Tutorials Point (I) Pvt. Ltd. All the content and graphics published in this e-book are the property of Tutorials Point (I) Pvt. Ltd. The user of this e-book is prohibited to reuse, retain, copy, distribute or republish any contents or a part of contents of this e-book in any manner without written consent of the publisher. We strive to update the contents of our website and tutorials as timely and as precisely as possible, however, the contents may contain inaccuracies or errors. Tutorials Point (I) Pvt. Ltd. provides no guarantee regarding the accuracy, timeliness or completeness of our website or its contents including this tutorial. If you discover any errors on our website or in this tutorial, please notify us at [email protected] ii Gensim Table of Contents About the Tutorial ..........................................................................................................................................
    [Show full text]
  • Deconstructing Word Embedding Models
    Deconstructing Word Embedding Models Koushik K. Varma Ashoka University, [email protected] Abstract – Analysis of Word Embedding Models through Analysis, one must be certain that all linguistic aspects of a a deconstructive approach reveals their several word are captured within the vector representations because shortcomings and inconsistencies. These include any discrepancies could be further amplified in practical instability of the vector representations, a distorted applications. While these vector representations have analogical reasoning, geometric incompatibility with widespread use in modern natural language processing, it is linguistic features, and the inconsistencies in the corpus unclear as to what degree they accurately encode the data. A new theoretical embedding model, ‘Derridian essence of language in its structural, contextual, cultural and Embedding,’ is proposed in this paper. Contemporary ethical assimilation. Surface evaluations on training-test embedding models are evaluated qualitatively in terms data provide efficiency measures for a specific of how adequate they are in relation to the capabilities of implementation but do not explicitly give insights on the a Derridian Embedding. factors causing the existing efficiency (or the lack there of). We henceforth present a deconstructive review of 1. INTRODUCTION contemporary word embeddings using available literature to clarify certain misconceptions and inconsistencies Saussure, in the Course in General Linguistics, concerning them. argued that there is no substance in language and that all language consists of differences. In language there are only forms, not substances, and by that he meant that all 1.1 DECONSTRUCTIVE APPROACH apparently substantive units of language are generated by other things that lie outside them, but these external Derrida, a post-structualist French philosopher, characteristics are actually internal to their make-up.
    [Show full text]
  • Vector Space Models for Words and Documents Statistical Natural Language Processing ELEC-E5550 Jan 30, 2019 Tiina Lindh-Knuutila D.Sc
    Vector space models for words and documents Statistical Natural Language Processing ELEC-E5550 Jan 30, 2019 Tiina Lindh-Knuutila D.Sc. (Tech) Lingsoft Language Services & Aalto University tiina.lindh-knuutila -at- aalto dot fi Material also by Mari-Sanna Paukkeri (mari-sanna.paukkeri -at- utopiaanalytics dot com) Today’s Agenda • Vector space models • word-document matrices • word vectors • stemming, weighting, dimensionality reduction • similarity measures • Count models vs. predictive models • Word2vec • Information retrieval (Briefly) • Course project details You shall know the word by the company it keeps • Language is symbolic in nature • Surface form is in an arbitrary relation with the meaning of the word • Hat vs. cat • One substitution: Levenshtein distance of 1 • Does not measure the semantic similarity of the words • Distributional semantics • Linguistic items which appear in similar contexts in large samples of language data tend to have similar meanings Firth, John R. 1957. A synopsis of linguistic theory 1930–1955. In Studies in linguistic analysis, 1–32. Oxford: Blackwell. George Miller and Walter Charles. 1991. Contextual correlates of semantic similarity. Language and Cognitive Processes, 6(1):1–28. Vector space models (VSM) • The use of a high-dimensional space of documents (or words) • Closeness in the vector space resembles closeness in the semantics or structure of the documents (depending on the features extracted). • Makes the use of data mining possible • Applications: – Document clustering/classification/… • Finding similar documents • Finding similar words – Word disambiguation – Information retrieval • Term discrimination: ranking keywords in the order of usefulness Vector space models (VSM) • Steps to build a vector space model 1. Preprocessing 2.
    [Show full text]
  • Word Embeddings: History & Behind the Scene
    Word Embeddings: History & Behind the Scene Alfan F. Wicaksono Information Retrieval Lab. Faculty of Computer Science Universitas Indonesia References • Bengio, Y., Ducharme, R., Vincent, P., & Janvin, C. (2003). A Neural Probabilistic Language Model. The Journal of Machine Learning Research, 3, 1137–1155. • Mikolov, T., Corrado, G., Chen, K., & Dean, J. (2013). Efficient Estimation of Word Representations in Vector Space. Proceedings of the International Conference on Learning Representations (ICLR 2013), 1–12 • Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Distributed Representations of Words and Phrases and their Compositionality. NIPS, 1–9. • Morin, F., & Bengio, Y. (2005). Hierarchical Probabilistic Neural Network Language Model. Aistats, 5. References Good weblogs for high-level understanding: • http://sebastianruder.com/word-embeddings-1/ • Sebastian Ruder. On word embeddings - Part 2: Approximating the Softmax. http://sebastianruder.com/word- embeddings-softmax • https://www.tensorflow.org/tutorials/word2vec • https://www.gavagai.se/blog/2015/09/30/a-brief-history-of- word-embeddings/ Some slides were also borrowed from From Dan Jurafsky’s course slide: Word Meaning and Similarity. Stanford University. Terminology • The term “Word Embedding” came from deep learning community • For computational linguistic community, they prefer “Distributional Semantic Model” • Other terms: – Distributed Representation – Semantic Vector Space – Word Space https://www.gavagai.se/blog/2015/09/30/a-brief-history-of-word-embeddings/ Before we learn Word Embeddings... Semantic Similarity • Word Similarity – Near-synonyms – “boat” and “ship”, “car” and “bicycle” • Word Relatedness – Can be related any way – Similar: “boat” and “ship” – Topical Similarity: • “boat” and “water” • “car” and “gasoline” Why Word Similarity? • Document Classification • Document Clustering • Language Modeling • Information Retrieval • ..
    [Show full text]
  • UNIVERSITY of CALIFORNIA, SAN DIEGO Bag-Of-Concepts As a Movie
    UNIVERSITY OF CALIFORNIA, SAN DIEGO Bag-of-Concepts as a Movie Genome and Representation A Thesis submitted in partial satisfaction of the requirements for the degree Master of Science in Computer Science by Colin Zhou Committee in charge: Professor Shlomo Dubnov, Chair Professor Sanjoy Dasgupta Professor Lawrence Saul 2016 The Thesis of Colin Zhou is approved, and it is acceptable in quality and form for publication on microfilm and electronically: Chair University of California, San Diego 2016 iii DEDICATION For Mom and Dad iv TABLE OF CONTENTS Signature Page........................................................................................................... iii Dedication.................................................................................................................. iv Table of Contents....................................................................................................... v List of Figures and Tables.......................................................................................... vi Abstract of the Thesis................................................................................................ vii Introduction................................................................................................................ 1 Chapter 1 Background............................................................................................... 3 1.1 Vector Space Models............................................................................... 3 1.1.1. Bag-of-Words........................................................................
    [Show full text]
  • Analysis of a Vector Space Model, Latent Semantic Indexing and Formal Concept Analysis for Information Retrieval
    BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES • Volume 12, No 1 Sofia • 2012 Analysis of a Vector Space Model, Latent Semantic Indexing and Formal Concept Analysis for Information Retrieval Ch. Aswani Kumar1, M. Radvansky2, J. Annapurna3 1School of Information Technology and Engineering, VIT University, Vellore, India 2VSB Technical University of Ostrava, Ostrava, Czech Republic 3School of Computing Science and Engineering, VIT University, Vellore, India Email: [email protected] Abstract: Latent Semantic Indexing (LSI), a variant of classical Vector Space Model (VSM), is an Information Retrieval (IR) model that attempts to capture the latent semantic relationship between the data items. Mathematical lattices, under the framework of Formal Concept Analysis (FCA), represent conceptual hierarchies in data and retrieve the information. However, both LSI and FCA use the data represented in the form of matrices. The objective of this paper is to systematically analyze VSM, LSI and FCA for the task of IR using standard and real life datasets. Keywords: Formal concept analysis, Information Retrieval, latent semantic indexing, vector space model. 1. Introduction Information Retrieval (IR) deals with the representation, storage, organization and access to information items. IR is highly iterative in human interactive process aimed at retrieving the documents that are related to users’ information needs. The human interaction consists in submitting information needed as a query, analyzing the ranked and retrieved documents, modifying the query and submitting iteratively until the actual documents related to the need are found or the search process is terminated by the user. Several techniques including a simple keyword to advanced NLP are available for developing IR systems.
    [Show full text]
  • Ontology Alignment Based on Word Embedding and Random Forest Classification
    Ontology alignment based on word embedding and random forest classification Ikechukwu Nkisi-Orji1[0000−0001−9734−9978], Nirmalie Wiratunga1, Stewart Massie1, Kit-Ying Hui1, and Rachel Heaven2 1 Robert Gordon University, Aberdeen, UK {i.o.nkisi-orji , n.wiratunga, s.massie, k.hui}@rgu.ac.uk 2 British Geological Survey, Nottingham, UK [email protected] Abstract. Ontology alignment is crucial for integrating heterogeneous data sources and forms an important component of the semantic web. Accordingly, several ontology alignment techniques have been proposed and used for discovering correspondences between the concepts (or enti- ties) of different ontologies. Most alignment techniques depend on string- based similarities which are unable to handle the vocabulary mismatch problem. Also, determining which similarity measures to use and how to effectively combine them in alignment systems are challenges that have persisted in this area. In this work, we introduce a random forest classifier approach for ontology alignment which relies on word embed- ding for determining a variety of semantic similarities features between concepts. Specifically, we combine string-based and semantic similarity measures to form feature vectors that are used by the classifier model to determine when concepts match. By harnessing background knowledge and relying on minimal information from the ontologies, our approach can handle knowledge-light ontological resources. It also eliminates the need for learning the aggregation weights of a composition of similarity measures. Experiments using Ontology Alignment Evaluation Initiative (OAEI) dataset and real-world ontologies highlight the utility of our approach and show that it can outperform state-of-the-art alignment systems. Keywords: Ontology alignment · Word embedding · Machine classifi- cation · Semantic web.
    [Show full text]
  • A Structured Vector Space Model for Word Meaning in Context
    A Structured Vector Space Model for Word Meaning in Context Katrin Erk Sebastian Pado´ Department of Linguistics Department of Linguistics University of Texas at Austin Stanford University [email protected] [email protected] Abstract all its possible usages (Landauer and Dumais, 1997; Lund and Burgess, 1996). Since the meaning of We address the task of computing vector space words can vary substantially between occurrences representations for the meaning of word oc- (e.g., for polysemous words), the next necessary step currences, which can vary widely according to context. This task is a crucial step towards a is to characterize the meaning of individual words in robust, vector-based compositional account of context. sentence meaning. We argue that existing mod- There have been several approaches in the liter- els for this task do not take syntactic structure ature (Smolensky, 1990; Schutze,¨ 1998; Kintsch, sufficiently into account. 2001; McDonald and Brew, 2004; Mitchell and La- We present a novel structured vector space pata, 2008) that compute meaning in context from model that addresses these issues by incorpo- lemma vectors. Most of these studies phrase the prob- rating the selectional preferences for words’ lem as one of vector composition: The meaning of a argument positions. This makes it possible to target occurrence a in context b is a single new vector integrate syntax into the computation of word c that is a function (for example, the centroid) of the meaning in context. In addition, the model per- forms at and above the state of the art for mod- vectors: c = a b.
    [Show full text]
  • Graph-Based Word Embedding Learning and Cross-Lingual Contextual Word Embedding Learning Zheng Zhang
    Explorations in Word Embeddings : graph-based word embedding learning and cross-lingual contextual word embedding learning Zheng Zhang To cite this version: Zheng Zhang. Explorations in Word Embeddings : graph-based word embedding learning and cross- lingual contextual word embedding learning. Computation and Language [cs.CL]. Université Paris Saclay (COmUE), 2019. English. NNT : 2019SACLS369. tel-02366013 HAL Id: tel-02366013 https://tel.archives-ouvertes.fr/tel-02366013 Submitted on 15 Nov 2019 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Explorations in word embeddings: S369 graph-based word embedding learning SACL and cross-lingual contextual word 9 embedding learning : 201 NNT Thèse de doctorat de l'Université Paris-Saclay préparée à Université Paris-Sud École doctorale n°580 Sciences et technologies de l’information et de la communication (STIC) Spécialité de doctorat : Informatique Thèse présentée et soutenue à Orsay, le 18/10/2019, par Zheng Zhang Composition du Jury : François Yvon Senior Researcher, LIMSI-CNRS Président Mathieu Lafourcade Associate Professor, Université de Montpellier Rapporteur Emmanuel Morin Professor, Université de Nantes Rapporteur Armand Joulin Research Scientist, Facebook Artificial Intelligence Research Examinateur Pierre Zweigenbaum Senior Researcher, LIMSI-CNRS Directeur de thèse Yue Ma Associate Professor, Université Paris-Sud Co-encadrant Acknowledgements Word embedding learning is a fast-moving domain.
    [Show full text]