Trading spaces : on the lore and limitations of latent semantic analysis

Two decades after its inception, Latent Semantic Analysis(LSA) has become part and parcel of every modern introduction to Information Retrieval. For any tool that matures so quickly, it is important to check its lore and limitations, or else stagnation will set in. We focus here on the three main aspects of LSA that are well accepted, and the gist of which can be summarized as follows: (1) that LSA recovers latent semantic factors underlying the document space, (2) that such can be accomplished through lossy compression of the document space by eliminating lexical noise, and (3) that the latter can best be achieved by Singular Value Decomposition. For each aspect we performed experiments analogous to those reported in the LSA literature and compared the evidence brought to bear in each case. On the negative side, we show that the above claims about LSA are much more limited than commonly believed. Even a simple example may show that LSA does not recover the optimal semantic factors as intended in the pedagogical example used in many LSA publications. Additionally, and remarkably deviating from LSA lore, LSA does not scale up well: the larger the document space, the more unlikely that LSA recovers an optimal set of semantic factors. On the positive side, we describe new algorithms to replace LSA (and more recent alternatives as pLSA, LDA, and kernel methods) by trading its l2 space for an l1 space, thereby guaranteeing an optimal set of semantic factors. These algorithms seem to salvage the spirit of LSA as we think it was initially conceived.

Effective web service discovery using a combination of a semantic model and a data mining technique

With the advent of Service Oriented Architecture, Web Services have gained tremendous popularity. Due to the availability of a large number of Web services, finding an appropriate Web service according to the requirement of the user is a challenge. This warrants the need to establish an effective and reliable process of Web service discovery. A considerable body of research has emerged to develop methods to improve the accuracy of Web service discovery to match the best service. The process of Web service discovery results in suggesting many individual services that partially fulfil the user’s interest. By considering the semantic relationships of words used in describing the services as well as the use of input and output parameters can lead to accurate Web service discovery. Appropriate linking of individual matched services should fully satisfy the requirements which the user is looking for. This research proposes to integrate a semantic model and a data mining technique to enhance the accuracy of Web service discovery. A novel three-phase Web service discovery methodology has been proposed. The first phase performs match-making to find semantically similar Web services for a user query. In order to perform semantic analysis on the content present in the Web service description language document, the support-based latent semantic kernel is constructed using an innovative concept of binning and merging on the large quantity of text documents covering diverse areas of domain of knowledge. The use of a generic latent semantic kernel constructed with a large number of terms helps to find the hidden meaning of the query terms which otherwise could not be found. Sometimes a single Web service is unable to fully satisfy the requirement of the user. In such cases, a composition of multiple inter-related Web services is presented to the user. The task of checking the possibility of linking multiple Web services is done in the second phase. Once the feasibility of linking Web services is checked, the objective is to provide the user with the best composition of Web services. In the link analysis phase, the Web services are modelled as nodes of a graph and an allpair shortest-path algorithm is applied to find the optimum path at the minimum cost for traversal. The third phase which is the system integration, integrates the results from the preceding two phases by using an original fusion algorithm in the fusion engine. Finally, the recommendation engine which is an integral part of the system integration phase makes the final recommendations including individual and composite Web services to the user. In order to evaluate the performance of the proposed method, extensive experimentation has been performed. Results of the proposed support-based semantic kernel method of Web service discovery are compared with the results of the standard keyword-based information-retrieval method and a clustering-based machine-learning method of Web service discovery. The proposed method outperforms both information-retrieval and machine-learning based methods. Experimental results and statistical analysis also show that the best Web services compositions are obtained by considering 10 to 15 Web services that are found in phase-I for linking. Empirical results also ascertain that the fusion engine boosts the accuracy of Web service discovery by combining the inputs from both the semantic analysis (phase-I) and the link analysis (phase-II) in a systematic fashion. Overall, the accuracy of Web service discovery with the proposed method shows a significant improvement over traditional discovery methods.

Quantitative approaches to content analysis : identifying conceptual drift across publication outlets

Unstructured text data, such as emails, blogs, contracts, academic publications, organizational documents, transcribed interviews, and even tweets, are important sources of data in Information Systems research. Various forms of qualitative analysis of the content of these data exist and have revealed important insights. Yet, to date, these analyses have been hampered by limitations of human coding of large data sets, and by bias due to human interpretation. In this paper, we compare and combine two quantitative analysis techniques to demonstrate the capabilities of computational analysis for content analysis of unstructured text. Specifically, we seek to demonstrate how two quantitative analytic methods, viz., Latent Semantic Analysis and data mining, can aid researchers in revealing core content topic areas in large (or small) data sets, and in visualizing how these concepts evolve, migrate, converge or diverge over time. We exemplify the complementary application of these techniques through an examination of a 25-year sample of abstracts from selected journals in Information Systems, Management, and Accounting disciplines. Through this work, we explore the capabilities of two computational techniques, and show how these techniques can be used to gather insights from a large corpus of unstructured text.

Semantics-based web service discovery using information retrieval techniques

This paper demonstrates an experimental study that examines the accuracy of various information retrieval techniques for Web service discovery. The main goal of this research is to evaluate algorithms for semantic web service discovery. The evaluation is comprehensively benchmarked using more than 1,700 real-world WSDL documents from INEX 2010 Web Service Discovery Track dataset. For automatic search, we successfully use Latent Semantic Analysis and BM25 to perform Web service discovery. Moreover, we provide linking analysis which automatically links possible atomic Web services to meet the complex requirements of users. Our fusion engine recommends a final result to users. Our experiments show that linking analysis can improve the overall performance of Web service discovery. We also find that keyword-based search can quickly return results but it has limitation of understanding users’ goals.

Detecting conversing groups of chatters : a model, algorithms, and tests

Chatrooms, for example Internet Relay Chat, are generally multi-user, multi-channel and multiserver chat-systems which run over the Internet and provide a protocol for real-time text-based conferencing between users all over the world. While a well-trained human observer is able to understand who is chatting with whom, there are no efficient and accurate automated tools to determine the groups of users conversing with each other. A precursor to analysing evolving cyber-social phenomena is to first determine what the conversations are and which groups of chatters are involved in each conversation. We consider this problem in this paper. We propose an algorithm to discover all groups of users that are engaged in conversation. Our algorithms are based on a statistical model of a chatroom that is founded on our experience with real chatrooms. Our approach does not require any semantic analysis of the conversations, rather it is based purely on the statistical information contained in the sequence of posts. We improve the accuracy by applying some graph algorithms to clean the statistical information. We present some experimental results which indicate that one can automatically determine the conversing groups in a chatroom, purely on the basis of statistical analysis.

A tag-based personalized item recommendation system using tensor modeling and topic model approaches

This research falls in the area of enhancing the quality of tag-based item recommendation systems. It aims to achieve this by employing a multi-dimensional user profile approach and by analyzing the semantic aspects of tags. Tag-based recommender systems have two characteristics that need to be carefully studied in order to build a reliable system. Firstly, the multi-dimensional correlation, called as tag assignment , should be appropriately modelled in order to create the user profiles [1]. Secondly, the semantics behind the tags should be considered properly as the flexibility with their design can cause semantic problems such as synonymy and polysemy [2]. This research proposes to address these two challenges for building a tag-based item recommendation system by employing tensor modeling as the multi-dimensional user profile approach, and the topic model as the semantic analysis approach. The first objective is to optimize the tensor model reconstruction and to improve the model performance in generating quality rec-ommendation. A novel Tensor-based Recommendation using Probabilistic Ranking (TRPR) method [3] has been developed. Results show this method to be scalable for large datasets and outperforming the benchmarking methods in terms of accuracy. The memory efficient loop implements the n-mode block-striped (matrix) product for tensor reconstruction as an approximation of the initial tensor. The probabilistic ranking calculates the probabil-ity of users to select candidate items using their tag preference list based on the entries generated from the reconstructed tensor. The second objective is to analyse the tag semantics and utilize the outcome in building the tensor model. This research proposes to investigate the problem using topic model approach to keep the tags nature as the “social vocabulary” [4]. For the tag assignment data, topics can be generated from the occurrences of tags given for an item. However there is only limited amount of tags availa-ble to represent items as collection of topics, since an item might have only been tagged by using several tags. Consequently, the generated topics might not able to represent the items appropriately. Furthermore, given that each tag can belong to any topics with various probability scores, the occurrence of tags cannot simply be mapped by the topics to build the tensor model. A standard weighting technique will not appropriately calculate the value of tagging activity since it will define the context of an item using a tag instead of a topic.

Deriving word association networks from text corpora

This article presents and evaluates a model to automatically derive word association networks from text corpora. Two aspects were evaluated: To what degree can corpus-based word association networks (CANs) approximate human word association networks with respect to (1) their ability to quantitatively predict word associations and (2) their structural network characteristics. Word association networks are the basis of the human mental lexicon. However, extracting such networks from human subjects is laborious, time consuming and thus necessarily limited in relation to the breadth of human vocabulary. Automatic derivation of word associations from text corpora would address these limitations. In both evaluations corpus-based processing provided vector representations for words. These representations were then employed to derive CANs using two measures: (1) the well known cosine metric, which is a symmetric measure, and (2) a new asymmetric measure computed from orthogonal vector projections. For both evaluations, the full set of 4068 free association networks (FANs) from the University of South Florida word association norms were used as baseline human data. Two corpus based models were benchmarked for comparison: a latent topic model and latent semantic analysis (LSA). We observed that CANs constructed using the asymmetric measure were slightly less effective than the topic model in quantitatively predicting free associates, and slightly better than LSA. The structural networks analysis revealed that CANs do approximate the FANs to an encouraging degree.

Empirical analysis of the effect of dimension reduction and word order on semantic vectors

The aim of this paper is to provide a comparison of various algorithms and parameters to build reduced semantic spaces. The effect of dimension reduction, the stability of the representation and the effect of word order are examined in the context of the five algorithms bearing on semantic vectors: Random projection (RP), singular value decom- position (SVD), non-negative matrix factorization (NMF), permutations and holographic reduced representations (HRR). The quality of semantic representation was tested by means of synonym finding task using the TOEFL test on the TASA corpus. Dimension reduction was found to improve the quality of semantic representation but it is hard to find the optimal parameter settings. Even though dimension reduction by RP was found to be more generally applicable than SVD, the semantic vectors produced by RP are somewhat unstable. The effect of encoding word order into the semantic vector representation via HRR did not lead to any increase in scores over vectors constructed from word co-occurrence in context information. In this regard, very small context windows resulted in better semantic vectors for the TOEFL test.

Inducing shades of meaning by matrix methods : a first step towards thematic analysis of opinion

This article explores two matrix methods to induce the ``shades of meaning" (SoM) of a word. A matrix representation of a word is computed from a corpus of traces based on the given word. Non-negative Matrix Factorisation (NMF) and Singular Value Decomposition (SVD) compute a set of vectors corresponding to a potential shade of meaning. The two methods were evaluated based on loss of conditional entropy with respect to two sets of manually tagged data. One set reflects concepts generally appearing in text, and the second set comprises words used for investigations into word sense disambiguation. Results show that for NMF consistently outperforms SVD for inducing both SoM of general concepts as well as word senses. The problem of inducing the shades of meaning of a word is more subtle than that of word sense induction and hence relevant to thematic analysis of opinion where nuances of opinion can arise.

Finding additional semantic entity information for search engines

Entity-oriented search has become an essential component of modern search engines. It focuses on retrieving a list of entities or information about the specific entities instead of documents. In this paper, we study the problem of finding entity related information, referred to as attribute-value pairs, that play a significant role in searching target entities. We propose a novel decomposition framework combining reduced relations and the discriminative model, Conditional Random Field (CRF), for automatically finding entity-related attribute-value pairs from free text documents. This decomposition framework allows us to locate potential text fragments and identify the hidden semantics, in the form of attribute-value pairs for user queries. Empirical analysis shows that the decomposition framework outperforms pattern-based approaches due to its capability of effective integration of syntactic and semantic features.

On Chinese Modal Particle A (啊) : A Pragmatic and Semantic Study

Chinese modal particles feature prominently in Chinese people’s daily use of the language, but their pragmatic and semantic functions are elusive as commonly recognised by Chinese linguists and teachers of Chinese as a foreign language. This book originates from an extensive and intensive empirical study of the Chinese modal particle a (啊), one of the most frequently used modal particles in Mandarin Chinese. In order to capture all the uses and the underlying meanings of the particle, the author transcribed the first 20 episodes, about 20 hours in length, of the popular Chinese TV drama series Kewang ‘Expectations’, which yielded a corpus data of more than 142’000 Chinese characters with a total of 1829 instances of the particle all used in meaningful communicative situations. Within its context of use, every single occurrence of the particle was analysed in terms of its pragmatic and semantic contributions to the hosting utterance. Upon this basis the core meanings were identified which were seen as constituting the modal nature of the particle.

Oscar Pistorius and the death of Reeva Steenkamp : mapping public opinions on Twitter using frame analysis and data mining

This paper uses innovative content analysis techniques to map how the death of Oscar Pistorius' girlfriend, Reeva Steenkamp, was framed on Twitter conversations. Around 1.5 million posts from a two-week timeframe are analyzed with a combination of syntactic and semantic methods. This analysis is grounded in the frame analysis perspective and is different than sentiment analysis. Instead of looking for explicit evaluations, such as “he is guilty” or “he is innocent”, we showcase through the results how opinions can be identified by complex articulations of more implicit symbolic devices such as examples and metaphors repeatedly mentioned. Different frames are adopted by users as more information about the case is revealed: from a more episodic one, highly used in the very beginning, to more systemic approaches, highlighting the association of the event with urban violence, gun control issues, and violence against women. A detailed timeline of the discussions is provided.

The semantic interference effect in the picture-word paradigm: An event-related fMRI study employing overt responses

We used event-related functional magnetic resonance imaging (fMRI) to investigate neural responses associated with the semantic interference (SI) effect in the picture-word task. Independent stage models of word production assume that the locus of the SI effect is at the conceptual processing level (Levelt et al. [1999]: Behav Brain Sci 22:1-75), whereas interactive models postulate that it occurs at phonological retrieval (Starreveld and La Heij [1996]: J Exp Psychol Learn Mem Cogn 22:896-918). In both types of model resolution of the SI effect occurs as a result of competitive, spreading activation without the involvement of inhibitory links. These assumptions were tested by randomly presenting participants with trials from semantically-related and lexical control distractor conditions and acquiring image volumes coincident with the estimated peak hemodynamic response for each trial. Overt vocalization of picture names occurred in the absence of scanner noise, allowing reaction time (RT) data to be collected. Analysis of the RT data confirmed the SI effect. Regions showing differential hemodynamic responses during the SI effect included the left mid section of the middle temporal gyrus, left posterior superior temporal gyrus, left anterior cingulate cortex, and bilateral orbitomedial prefrontal cortex. Additional responses were observed in the frontal eye fields, left inferior parietal lobule, and right anterior temporal and occipital cortex. The results are interpreted as indirectly supporting interactive models that allow spreading activation between both conceptual processing and phonological retrieval levels of word production. In addition, the data confirm that selective attention/response suppression has a role in resolving the SI effect similar to the way in which Stroop interference is resolved. We conclude that neuroimaging studies can provide information about the neuroanatomical organization of the lexical system that may prove useful for constraining theoretical models of word production.

Neural activity associated with semantic versus phonological anomia treatments in aphasia

Naming impairments in aphasia are typically targeted using semantic and/or phonologically based tasks. However, it is not known whether these treatments have different neural mechanisms. Eight participants with aphasia received twelve treatment sessions using an alternating treatment design, with fMRI scans pre- and post-treatment. Half the sessions employed Phonological Components Analysis (PCA), and half the sessions employed Semantic Feature Analysis (SFA). Pre-treatment activity in the left caudate correlated with greater immediate treatment success for items treated with SFA, whereas recruitment of the left supramarginal gyrus and right precuneus post-treatment correlated with greater immediate treatment success for items treated with PCA. The results support previous studies that have found greater treatment outcome to be associated with activity in predominantly left hemisphere regions, and suggest that different mechanisms may be engaged dependent on the type of treatment employed.