DOCSLIB.ORG

Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Kjell, Oscar

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Kjell, Oscar Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Kjell, Oscar 2018 Document Version: Publisher's PDF, also known as Version of record Link to publication Citation for published version (APA): Kjell, O. (2018). Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales. Lund University. Total number of authors: 1 General rights Unless other specific re-use rights are stated the following general rights apply: Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal Read more about Creative commons licenses: https://creativecommons.org/licenses/ Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. LUND UNIVERSITY PO Box 117 221 00 Lund +46 46-222 00 00 OSCAR KJELL 1 1 1 0 1 0 0 1 0 minds 1 0 0 0 1 1 0 1 concord 0 0 0 family 1mutual peace 0 0 2 0 1 1 life 1 0 1 sympathy food 0 1 1 0 0 love 0 AN ECOLABEL 3041 0903 0 ambition 1 1 0 1 1 1 1 Conceptualizing and Measuring Using Statistical Well-Being Semantics and Numerical Rating Scales 1 balance 0 1 1 compromise 0 accord 0 0 0 1 0 1 0 1 0 1 1 1 1 1 1 α 1 1 enjoyment expectations 0 car 0 0 1 0 0 forest 4 important 0 Conceptualizing and Measuring Well-Being Using α 1 0 0 1 0 0 joy 1 0 1 ryck, Lund 2018 NORDIC SW 1 Statistical Semantics and Numerical Rating Scales 5 1 0 0 1 friends 0 nature 1 creative 0 1 0 1 OSCAR KJELL 1 1 animals DEPARTMENT OF PSYCHOLOGY | FACULTY OF SOCIAL SCIENCES | LUND UNIVERSITY 0 1 Printed by Media-T 0 1 content reward wealth 1 sex 0 1 1 0 1 1 0 0 1 1 1 winning 0 knowledge 0 0 1 7 along 1 1 1 0 0 tranquillity 0 1 1 1 1 0 together amicability 0 1 0 0 1 0 1 0 1 0 pleased gratified own 1 0 0 0 0 0 0 1 0 0 1 helping 1 cooperation ω broaden 1 children 0 0 0 0 1 satisfaction 0 0 1 1 0 0 1 study 0 0 0 0 1 yoga 1 0 1 1 1 0 meditation calm 1 1 1 0 0 0 0 0 1 6 1 harmony0 0 1 1 0 privilege 1 0 6 1 0 1 0 1 1 1 happy 0 job 1 1 education 0 0 0 0 0 0 1 1 0 0 0 0 inner 0 1 0 1 1 success relaxation 0 0 unity peaceful goals 0 0 0 song 1 1 kids 0 0 0 1 1 0 0 good 0 fulfilled 1 1 faith 4 1 1 0 1 1 forgiveness 1 1 1 contentment 1 empathy consistency 0 1 career 0 1 1 1 0 1 0 1 0 0 1 1 7 music agreement 0 0 1 1 happiness money college 0 0 0 0 0 promotion 3 0 affinity 1 0 1 1 1 0 1 0 1 1 0 1 0 0 gentle 0 1 house 0 0 0 1 organisation 0 1 0 fight 1 2 0 1 equilibrium 0 1 ω pleasure 0 have 0 0 energy 1 achievement 1 0 0 0 5 0 0 1 0 1 balanced 1 0 tolerant 0 0 1 1 1 0 0 0 3 revenge 0 0 entertainment correspondence 1 1 1 friendly 1 1 0 0 openness understanding 1 1 0 1 1 1 0 1 535928 1 0 0 0 Lund University, Faculty of Social Sciences 0 789177 1 1 ISBN 978-91-7753-592-8 9 1 Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Oscar N.E. Kjell DOCTORAL DISSERTATION by due permission of the Faculty of Social Sciences, Lund University, Sweden. To be defended in Kulturen’s auditorium, Lund, the 13th of April 2018 at 13.15. Faculty opponent H. Andrew Schwartz, Stony Brook University 1 Organization Document name LUND UNIVERSITY Date of issue Author(s): Oscar N.E. Kjell Sponsoring organization Title and subtitle: Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Abstract How to define and measure individuals’ well-being is important, as this has an impact on both research and society at large. This thesis concerns how to define and measure the self-reported well-being of individuals, which involves both theorizing as well as developing and applying empirical and statistical methods in order to gain a better understanding of well-being. The first paper critically reviews the literature on well-being. It identifies an individualistic bias in current approaches and accompanying measures related to well-being and happiness; for example, through an over-emphasis on the importance of self-centered aspects of well-being (e.g., the unprecedented focus on satisfaction with life) whilst disregarding the importance of harmony in life, interconnectedness and psychological balance in relation to well- being. It is also discussed how closed-ended well-being measures impose the researchers’ values and limit the ability of respondents to express themselves in regard to their perceived well-being. The second paper addresses concerns regarding this individualistic bias by developing the harmony in life scale, which focuses on interconnectedness and psychological balance. In addition, an open-ended approach is developed in the paper, allowing individuals to freely describe their pursuit of well-being by means of open-ended responses analyzed using statistical semantics (including techniques from artificial intelligence such as natural language processing and machine learning). The results show that the harmony in life scale and the traditional satisfaction with life scale form a two-factor model of well-being, where the harmony in life scale explains more unique variance in measures of psychological well-being, stress, depression and anxiety, but not happiness. It is further demonstrated that participants describe their pursuit of harmony in life using words related to interconnectedness (including words such as: peace, balance, cooperation), whereas they describe their pursuit of satisfaction with life using words related to independence (including words such as: money, achievement, fulfillment). It is concluded that the harmony in life scale complements the satisfaction with life scale for a more comprehensive understanding of subjective well-being. The third paper focuses on developing and evaluating a method for measuring and describing psychological constructs using open-ended questions analyzed by means of statistical semantics rather than closed-ended numerical rating scales. This semantic measures approach is tested and compared with traditional rating scales in nine studies, including two different paradigms involving reports regarding objective stimuli (i.e., the evaluation of facial expressions) and reports regarding subjective states (i.e., the self-reporting of harmony in life, satisfaction with life, depression and worry). The results indicate that semantic measures encompass higher, or competitive, levels of reliability and validity compared to traditional numerical rating scales. In addition, semantic measures appear to be better suited for differentiating between psychological constructs, such as harmony in life versus satisfaction with life as well as depression versus worry. In this thesis, the findings from these three papers are elaborated and integrated into two independent perspectives. The first perspective focuses on the theoretical and empirical differences between harmony in life and satisfaction with life within a context of societal and national progress. It is concluded that harmony in life complements satisfaction with life. The second perspective focuses on the open-ended, statistical semantics approach. It is proposed that statistical semantics may beneficially be used more widely as a research tool within psychological research. Key words: Well-being, Harmony in life, Satisfaction with life, Statistical semantics, Latent semantic analysis. Classification system and/or index terms (if any) Supplementary bibliographical information Language: English ISSN and key title ISBN (Print): 978-91-7753-592-8 ISBN (Electronic): 978-91-7753-593-5 Recipient’s notes Number of pages Price Security classification I, the undersigned, being the copyright owner of the abstract of the above-mentioned dissertation, hereby grant to all reference sources permission to publish and disseminate the abstract of the above-mentioned dissertation. Signature Date 2018-02-26 2 Conceptualizing and Measuring Well-Being Using Statistical Semantics and Numerical Rating Scales Oscar N.E. Kjell 3 Coverphoto by Oscar Kjell Copyright Paper I American Psychological Association Copyright Paper II Springer Copyright Paper II and Thesis Oscar N.E. Kjell Faculty of Social Sciences Department of Psychology ISBN (Print): 978-91-7753-592-8 ISBN (Electronic): 978-91-7753-593-5 Printed in Sweden by Media-Tryck, Lund University Lund 2018 4 5 Table of Contents Acknowledgement .......................................................................................... 8 Abstract ........................................................................................................ 10 Sammanfattning ........................................................................................... 12 List of Papers ................................................................................................ 15 List of Abbreviations .................................................................................... 16 Introduction .......................................................................................................... 17 Overview of the Papers ................................................................................ 18 Format and Aims of the Thesis .................................................................... 21 References .................................................................................................... 23 Perspective I Harmony in Life Complements Satisfaction with Life in Measuring Subjective Well-Being for National Progress* ........................................................................ 25 Abstract ........................................................................................................ 25 National Progress Indicators and Well-Being .............................................
Load more

Recommended publications
  • Short-Text Clustering Using Statistical Semantics
    Short-Text Clustering using Statistical Semantics Sepideh Seifzadeh Ahmed K. Farahat Mohamed S. Kamel University of Waterloo University of Waterloo University of Waterloo Waterloo, Ontario, Canada. Waterloo, Ontario, Canada. Waterloo, Ontario, Canada. N2L 3G1 N2L 3G1 N2L 3G1 [email protected] [email protected] [email protected] Fakhri Karray University of Waterloo Waterloo, Ontario, Canada. N2L 3G1 [email protected] ABSTRACT 1. INTRODUCTION Short documents are typically represented by very sparse In social media, users usually post short texts. Twitter vectors, in the space of terms. In this case, traditional limits the length of each Tweet to 140 characters; therefore, techniques for calculating text similarity results in measures developing data mining techniques to handle the large vol- which are very close to zero, since documents even the very ume of short texts has become an important goal [1]. Text similar ones have a very few or mostly no terms in common. document clustering has been widely used to organize doc- In order to alleviate this limitation, the representation of ument databases and discover similarity and topics among short-text segments should be enriched by incorporating in- documents. Short text clustering is more challenging than formation about correlation between terms. In other words, regular text clustering; due to the sparsity and noise, they if two short segments do not have any common words, but provide very few contextual clues for applying traditional terms from the first segment appear frequently with terms data mining techniques [2]; therefore, short documents re- from the second segment in other documents, this means quire different or more adapted approaches.
    [Show full text]
  • Distributional Semantics
    Distributional semantics Distributional semantics is a research area that devel- by populating the vectors with information on which text ops and studies theories and methods for quantifying regions the linguistic items occur in; paradigmatic sim- and categorizing semantic similarities between linguis- ilarities can be extracted by populating the vectors with tic items based on their distributional properties in large information on which other linguistic items the items co- samples of language data. The basic idea of distributional occur with. Note that the latter type of vectors can also semantics can be summed up in the so-called Distribu- be used to extract syntagmatic similarities by looking at tional hypothesis: linguistic items with similar distributions the individual vector components. have similar meanings. The basic idea of a correlation between distributional and semantic similarity can be operationalized in many dif- ferent ways. There is a rich variety of computational 1 Distributional hypothesis models implementing distributional semantics, includ- ing latent semantic analysis (LSA),[8] Hyperspace Ana- The distributional hypothesis in linguistics is derived logue to Language (HAL), syntax- or dependency-based from the semantic theory of language usage, i.e. words models,[9] random indexing, semantic folding[10] and var- that are used and occur in the same contexts tend to ious variants of the topic model. [1] purport similar meanings. The underlying idea that “a Distributional semantic models differ primarily with re- word is characterized by the company it keeps” was pop- spect to the following parameters: ularized by Firth.[2] The Distributional Hypothesis is the basis for statistical semantics. Although the Distribu- • tional Hypothesis originated in linguistics,[3] it is now re- Context type (text regions vs.
    [Show full text]
  • Larkc FP7 – 215535 D1.2.2 Improved Operational Framework
    LarKC The Large Knowledge Collider a platform for large scale integrated reasoning and Web-search FP7 { 215535 D1.2.2 Improved Operational Framework Coordinator: Gaston Tagni With contributions from: Gaston Tagni (VUA), Zhisheng Huang (VUA), Stefan Schlobach (VUA), Annette ten Teije (VUA), Frank van Harmelen (VUA), Barry Bishop (UIBK), Florian Fischer (UIBK), Vassil Momtchev (Ontotext), Yi Zeng (WICI), Yan Wang (WICI), Yi Huang (Siemens), Georgina Gallizo (HLRS), Matthias Assel (HLRS), Jose Quesada (MPI) Quality Assesor: Michael Witbrock (CycEur) Quality Controller: Reto Krummenacher (UIBK) Document Identifier: LarKC/2008/D1.2.2 Class Deliverable: LarKC EU-IST-2008-215535 Version: version 1.0.0 Date: March 30, 2010 State: final Distribution: public FP7 { 215535 Deliverable 1.2.2 Executive Summary This deliverable is the second of a series of three deliverables aimed at defining an Operational Framework for scalable reasoning in LarKC and as such is a continu- ation of the work reported in Deliverable D1.2.1 - Initial Operational Framework (M7). The main contributions of this document are: First, a detailed analysis and discussion of different plug-ins currently being developed in LarKC in the context of the different technical work packages along with a discussion of several re-use possibilities between plug-ins and their sub-components. The second con- tribution is the specification of a series of design patterns aimed at supporting the development of plug-ins and other components in LarKC to achieve reasoning at Web-scale. 1 FP7 { 215535
    [Show full text]
  • Semantic Excel: an Introduction to a User-Friendly Online Software Application for Statistical Analyses of Text Data
    Semantic Excel: An Introduction to a User-Friendly Online Software Application for Statistical Analyses of Text Data Sikström1* S., Kjell1 O.N.E. & Kjell1 K. 1Department of Psychology, Lund University, Sweden *Correspondence to: [email protected] Acknowledges. We would like to thank Igor Marchetti for suggesting improvements on earlier drafts of this article. 1 Abstract Semantic Excel (www.semanticexcel.com) is an online software application with a simple, yet powerful interface enabling users to perform statistical analyses on texts. The purpose of this software is to facilitate statistical testing based on words, rather than numbers. The software comes with semantic representations, or an ordered set of numbers describing the semantic similarity between words/texts that are generated from Latent Semantic Analysis. These semantic representations are based on large datasets from Google N-grams for a dozen of the most commonly used languages in the world. This small-by-big data approach enables users to conduct analyses of small data that is enhanced by semantic knowledge from big data. First, we describe the theoretical foundation of these representations. Then we show the practical steps involved in carrying out statistical calculation using these semantic representations in Semantic Excel. This includes calculation of semantic similarity scores (i.e., computing a score describing the semantic similarity between two words/texts), semantic t-tests (i.e., statistically test whether two sets of words/texts differ in meaning), semantic-numeric correlations (i.e., statistically examine the relationship between words/texts and a numeric variable) and semantic predictions (i.e., using statistically trained models to predict numerical values from words/texts).
    [Show full text]
  • Neural Methods Towards Concept Discovery from Text Via Knowledge Transfer
    Neural Methods Towards Concept Discovery from Text via Knowledge Transfer DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Manirupa Das, B.E., M.S. Graduate Program in Computer Science and Engineering The Ohio State University 2019 Dissertation Committee: Prof Rajiv Ramnath, Advisor Prof Eric Fosler-Lussier, Advisor Prof Huan Sun c Copyright by Manirupa Das 2019 ABSTRACT Novel contexts, consisting of a set of terms referring to one or more concepts, often arise in real-world querying scenarios such as; a complex search query into a document retrieval system or a nuanced subjective natural language question. The concepts in these queries may not directly refer to entities or canonical concept forms occurring in any fact-based or rule-based knowledge source such as a knowledge base or on- tology. Thus, in addressing the complex information needs expressed by such novel contexts, systems using only such sources can fall short. Moreover, hidden associa- tions meaningful in the current context, may not exist in a single document, but in a collection, between matching candidate concepts having different surface realizations, via alternate lexical forms. These may refer to underlying latent concepts, i.e., exist- ing or conceived concepts or semantic classes that are accessible only via their surface forms. Inferring these latent concept associations in an implicit manner, by transfer- ring knowledge from the same domain { within a collection, or from across domains (different collections), can potentially better address such novel contexts. Thus latent concept associations may act as a proxy for a novel context.
    [Show full text]
  • Arxiv:1003.1141V1 [Cs.CL]
    Journal of Artificial Intelligence Research 37 (2010) 141-188 Submitted 10/09; published 02/10 From Frequency to Meaning: Vector Space Models of Semantics Peter D. Turney [email protected] National Research Council Canada Ottawa, Ontario, Canada, K1A 0R6 Patrick Pantel [email protected] Yahoo! Labs Sunnyvale, CA, 94089, USA Abstract Computers understand very little of the meaning of human language. This profoundly limits our ability to give instructions to computers, the ability of computers to explain their actions to us, and the ability of computers to analyse and process text. Vector space models (VSMs) of semantics are beginning to address these limits. This paper surveys the use of VSMs for semantic processing of text. We organize the literature on VSMs according to the structure of the matrix in a VSM. There are currently three broad classes of VSMs, based on term–document, word–context, and pair–pattern matrices, yielding three classes of applications. We survey a broad range of applications in these three categories and we take a detailed look at a specific open source project in each category. Our goal in this survey is to show the breadth of applications of VSMs for semantics, to provide a new perspective on VSMs for those who are already familiar with the area, and to provide pointers into the literature for those who are less familiar with the field. 1. Introduction One of the biggest obstacles to making full use of the power of computers is that they currently understand very little of the meaning of human language.
    [Show full text]
  • Parallel Search Through Statistical Semantic Spaces Leveraging Linked Open Data
    Parallel Search Through Statistical Semantic Spaces Leveraging Linked Open Data Alexey Cheptsov High-Performance Computing Center Stuttgart, Nobelstr. 19, 70569 Stuttgart, Germany [email protected] Abstract. With billions of triples in the Linked Open Data cloud, which continues to grow exponentially, challenging tasks start to emerge related to the exploitation and reasoning of Web data. A considerable amount of work has been done in the area of using Information Retrieval (IR) methods to address these problems. However, although applied models work on the Web scale, they downgrade the semantics contained in an RDF graph by observing each physical resource as a ’bag of words (URIs/literals)’. Distributional statistic methods can address this problem by capturing the structure of the graph more efficiently. However, these methods are computationally expensive. In this paper, we describe the parallelization algorithm of one such method (Random Indexing) based on the Message-Passing Interface technology. Our evaluation results show super linear improvement Keywords: Statistical Semantics, Random Indexing, Parallelization, High Performance Computing, Message-Passing Interface. 1 Introduction We live in a big data world, which is already estimated to be of the size of several Zetta (10 21 ) Bytes. However, the most considerable growth has seen the linked (open) data domain. Recent years have seen a tremendous increase of structured data on the Web with public sectors such as UK and USA governments opening their data to public (e.g., the U.S.’s data.gov initiative [1]), and encouraging others to build useful applications. At the same time, Linked Open Data (LOD) [2] project continues stimulating creation, publication and interlinking the RDF graphs with those already in the LOD cloud.
    [Show full text]
  • SCALABILITY of SEMANTIC ANALYSIS in NATURAL LANGUAGE PROCESSING Rndr. Radim ˇreh˚Urek Ph.D. Thesis Faculty of Informatics Masa
    SCALABILITYOFSEMANTICANALYSIS INNATURALLANGUAGEPROCESSING RNDr. Radim Rehˇ ˚uˇrek Û¡¢£¤¥¦§¨ª«¬­Æ°±²³´µ·¸¹º»¼½¾¿Ý Ph.D. Thesis Faculty of Informatics Masaryk University May, 2011 Thesis Supervisor Assoc. Prof. PhDr. Karel Pala, CSc. Thesis Advisor Assoc. Prof. RNDr. Petr Sojka, Ph.D. ABSTRACT Data mining applications that work over input of very large scale (web- scale problems) pose challenges that are new and exciting both academi- cally and commercially. Any web-scale algorithm must be robust (deal- ing gracefully with the inevitable data noise), scalable (capable of ef- ficiently processing large input) and reasonably automated (as human intervention is very costly and often impossible on such scales). This thesis consists of two parts. In the first part, I explore scalabil- ity of methods that derive a semantic representation of plain text docu- ments. The focus will be entirely on unsupervised techniques, that is, on methods that do not make use of manually annotated resources or hu- man input. I develop and present scalable algorithms for Latent Seman- tic Analysis (LSA) and Latent Dirichlet Allocation (LDA), two general- purpose statistical methods for semantic analysis that serve as building blocks for more concrete, applied algorithms. Scalability is achieved by building the semantic models in a constant amount of memory and dis- tributing the computation over a cluster of autonomous computers, con- nected by a high-latency network. In addition, the novel LSA training algorithm operates in a single pass over the training data, allowing con- tinuous online training over infinite-sized training streams. The second part of the thesis deals with possible applications of these general semantic algorithms.
    [Show full text]
  • Arxiv:1702.04946V1 [Cs.DL] 16 Feb 2017
    Noname manuscript No. (will be inserted by the editor) Clustering articles based on semantic similarity Shenghui Wang · Rob Koopman Received: date / Accepted: date Abstract Document clustering is generally the first step for topic identifi- cation. Since many clustering methods operate on the similarities between documents, it is important to build representations of these documents which keep their semantics as much as possible and are also suitable for efficient sim- ilarity calculation. As we describe in [15], the metadata of articles in the Astro dataset contribute to a semantic matrix, which uses a vector space to capture the semantics of entities derived from these articles and consequently supports the contextual exploration of these entities in LittleAriadne. However, this se- mantic matrix does not allow to calculate similarities between articles directly. In this paper, we will describe in detail how we build a semantic representa- tion for an article from the entities that are associated with it. Base on such semantic representations of articles, we apply two standard clustering meth- ods, K-Means and the Louvain community detection algorithm, which leads to our two clustering solutions labelled as OCLC-31 (standing for K-Means) and OCLC-Louvain (standing for Louvain). In this paper, we will give the implementation details and a basic comparison with other clustering solutions that are reported in this special issue. Keywords semantic indexing · clustering · visualisation · K-Means · Louvain community detection R. Koopman OCLC Research, Schipholweg 99, Leiden, The Netherlands Tel.: +31 71 524 6500 E-mail: [email protected] S. Wang arXiv:1702.04946v1 [cs.DL] 16 Feb 2017 OCLC Research, Schipholweg 99, Leiden, The Netherlands Tel.: +31 71 524 6500 E-mail: [email protected] 2 Shenghui Wang, Rob Koopman 1 Introduction Topics, sub-fields, specialities build the core in the self-organised process of scientific knowledge production [6].
    [Show full text]
  • Statistical Semantics in Context Space: Amrita CEN@Author Profiling
    Statistical Semantics in Context Space: Amrita CEN@Author Profiling Barathi Ganesh HB1, Anand Kumar M2, and Soman KP2 1Artificial Intelligence Practice, Tata Consultancy Services, Kochi - 682 042, Kerala, India 2Center for Computational Engineering and Networking, Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham, Amrita University, India [email protected],[email protected],[email protected] Abstract. Languages shared by people differs due to diversity in their ethnicity, socioeconomic status, gender, language, religion, sexual orien- tation, geographical area, accents, pronunciation and word usages. This eventually fall into hypothesis that they follow unknown hidden pattern. By using this hypothesis, determining the class of a person such as age, gender, their personality and nativity has multiple applications in social media, forensic science, marketing analysis, e-commerce and e-security. This work advances the research on author profiling much further by overcoming existing language dependent, domain dependent and lexicon based author profiling methods by finding user's sociolect aspects based on authors statistical pattern of semantics in context space. It proves to be a domain and language independent method in Author Profiling by nearing constant performance over English, Dutch and Spanish corpus. Keywords: Author Profiling, Context Space, Distributional Represen- tation 1 Introduction The amount of language sharing through Internet is prevalent due to the rapid growth of the social media resources like Facebook, Twitter, LinkedIn, Pinterest and chat resources like Hike, Whatapp, Wechat [1]. This positive growth ensures and encourages the recommendation and Internet marketing among users on a particular resource. It has been used in business organization for the marketing, market analysis, advertising and connecting with customers [2].
    [Show full text]
  • Understand Functionality and Dimensionality of Vector Embeddings: the Distributional Hypothesis, the Pairwise Inner Product Loss and Its Bias-Variance Trade-Off
    Understand Functionality and Dimensionality of Vector Embeddings: the Distributional Hypothesis, the Pairwise Inner Product Loss and Its Bias-Variance Trade-off Zi Yin [email protected] Department of Electrical Engineering Stanford University Stanford, CA 94305, USA May 22, 2018 Abstract Vector embedding is a foundational building block of many deep learning models, especially in natural language processing. In this paper, we present a theoretical framework for under- standing the effect of dimensionality on vector embeddings. We observe that the distributional hypothesis, a governing principle of statistical semantics, requires a natural unitary-invariance for vector embeddings. Motivated by the unitary-invariance observation, we propose the Pair- wise Inner Product (PIP) loss, a unitary-invariant metric on the similarity between two em- beddings. We demonstrate that the PIP loss captures the difference in functionality between embeddings, and that the PIP loss is tightly connect with two basic properties of vector embed- dings, namely similarity and compositionality. By formulating the embedding training process as matrix factorization with noise, we reveal a fundamental bias-variance trade-off between the signal spectrum and noise power in the dimensionality selection process. This bias-variance trade-off sheds light on many empirical observations which have not been thoroughly explained, for example the existence of an optimal dimensionality. Moreover, we discover two new results about vector embeddings, namely their robustness against over-parametrization and their for- ward stability. The bias-variance trade-off of the PIP loss explicitly answers the fundamental open problem of dimensionality selection for vector embeddings. 1 Introduction arXiv:1803.00502v4 [stat.ML] 21 May 2018 Vector space models (VSMs) are extremely useful and versatile tools that serve as keys to many fun- damental problems in numerous research areas (Turney and Pantel, 2010).
    [Show full text]
  • The Construction of Meaning
    The Construction of Meaning Walter Kintsch & Praful Mangalath University of Colorado∗ We argue that word meanings are not stored in a mental lexicon but are generated in the context of working memory from long-term memory traces which record our experience with words. Current statistical models of semantics, such as LSA and the Topic Model, describe what is stored in long-term memory. The CI-2 model describes how this information is used to construct sentence meanings. This model is a dual-memory model, in that it distinguishes between a gist level and an explicit level. It also incorporates syntactic information about how words are used, derived from dependency grammar. The construction of meaning is conceptualized as feature sampling from the explicit memory traces, with the constraint that the sampling must be contextually relevant both semantically and syntactically. Semantic relevance is achieved by sampling topically relevant features; local syntactic constraints as expressed by dependency relations ensure syntactic relevance. In the present paper, we are concerned with how meaning can be inferred from the analysis of large linguistic corpora. Specifically, our goal is to present a model of how sentence meanings are constructed as opposed to word meanings per se. To do so, we need to combine semantic and syntactic information about words stored in long-term 1 memory with local information about the sentence context. We first review models that extract latent semantic information from linguistic corpora and show how that information can be contextualized in working memory. We then present a model that uses not only semantic information at the gist level, but also explicit information about the actual patterns of word use to arrive at sentence interpretations.
    [Show full text]