Semantic Vector Space Research Articles

Combining content with user preferences for non-fiction multimedia recommendation: a study on TED lectures

This paper introduces a new dataset and compares several methods for the recommendation of non-fiction audio visual material, namely lectures from the TED website. The TED dataset contains 1,149 talks and 69,023 profiles of users, who have made more than 100,000 ratings and 200,000 comments. The corresponding metadata, which we make available, can be used for training and testing generic or personalized recommender systems. We define content-based, collaborative, and combined recommendation methods for TED lectures and use cross-validation to select the best parameters of keyword-based (TFIDF) and semantic vector space-based methods (LSI, LDA, RP, and ESA). We compare these methods on a personalized recommendation task in two settings, a cold-start and a non-cold-start one. In the cold-start setting, semantic vector spaces perform better than keywords. In the non-cold-start setting, where collaborative information can be exploited, content-based methods are outperformed by collaborative filtering ones, but the proposed combined method shows acceptable performances, and can be used in both settings. For the generic recommendation task, LSI and RP again outperform TF-IDF.

A novel semantic level text classification by combining NLP and Thesaurus concepts

Text categorization (also known as text classification or topic spotting) is the task of automatically sorting a set of documents into categories from a predefined set. Automated text classification is attractive because it frees organizations from the need of manually organizing document bases, but it can be too expensive or simply not feasible given the time constraints of the application or the number of documents involved. In the previous approaches only the Wikipedia concepts related to terms in syntactic level are used to represent document in semantic level. This paper proposes a new approach to represent semantic level with the use of Word Net. The semantic weight of terms related to the concepts from Wikipedia and Word Net are used to represent semantic information. The semantic vector space model of terms by combining the Word Net and Wikipedia is being further improved the classification accuracy of the Text classification. Because of, two different concept extractor are gives the concepts related to the terms in the syntactic level o find the better concept vector space for documents. So we obtain the improved classification by using this approach. In this study the classification framework are presented. In classification framework, the primary information is effectively kept and the noise is reduced by compressing the original information, so that this framework can guarantee the quality of the input of all classifiers. This proposed method can help to further improve the performance of classification framework by introducing Wikipedia with Word Net. We find that the proposed

Utilizing Conceptual Indexing to Enhance the Effectiveness of Vector Space Model

One of the main purposes of the semantic Web is to improve the retrieval performance of search systems. Unlike keyword based search systems, the semantic search systems aim to discover pages related to the query's concepts rather than merely collecting all pages instantiating its keywords. To that end, the concepts must be defined to be used as a semantic index instead of the traditional lexical one. In fact, The Arabic language is still far from being semantically searchable. Therefore, this paper proposed a model that exploits the Universal Word Net ontology for producing an Arabic Concepts-Space to be used as the index of Semantic Vector Space Model. The Vector Space Model is one of the most common information retrieval models due to its capability of expressing the documents' structure. However, like all keyword-based search systems, its sensitivity to the query's keywords reduces its retrieval effectiveness. The proposed model allows the VSM to represent Arabic documents by their topic, and thus classify them semantically. This, consequently, enhances the retrieval effectiveness of the search system.

Semantic smoothing for text clustering

In this paper we present a new semantic smoothing vector space kernel (S-VSM) for text documents clustering. In the suggested approach semantic relatedness between words is used to smooth the similarity and the representation of text documents. The basic hypothesis examined is that considering semantic relatedness between two text documents may improve the performance of the text document clustering task. For our experimental evaluation we analyze the performance of several semantic relatedness measures when embedded in the proposed (S-VSM) and present results with respect to different experimental conditions, such as: (i) the datasets used, (ii) the underlying knowledge sources of the utilized measures, and (iii) the clustering algorithms employed. To the best of our knowledge, the current study is the first to systematically compare, analyze and evaluate the impact of semantic smoothing in text clustering based on ‘wisdom of linguists’, e.g., WordNets, ‘wisdom of crowds’, e.g., Wikipedia, and ‘wisdom of corpora’, e.g., large text corpora represented with the traditional Bag of Words (BoW) model. Three semantic relatedness measures for text are considered; two knowledge-based (Omiotis[1] that uses WordNet, and WLM[2] that uses Wikipedia), and one corpus-based (PMI[3] trained on a semantically tagged SemCor version). For the comparison of different experimental conditions we use the BCubed F-Measure evaluation metric which satisfies all formal constraints of good quality cluster. The experimental results show that the clustering performance based on the S-VSM is better compared to the traditional VSM model and compares favorably against the standard GVSM kernel which uses word co-occurrences to compute the latent similarities between document terms.

A Cloud Support Vector Machine Model Based on Image Semantics

In allusion to randomness and fuzziness of digital image semantic, we propose a new semantic representation of digital image based on cloud model and construct a semantic vector space. In this space, semantic classifications of digital images are completed by calculating the semantic class certainty degree (SCCD). In addition, we propose cloud support vector machine based on image semantics (CSVM-IS) model. Experimental results show that CSVM-IS can accomplish target classification and has good classification accuracy.

Scale exhaustivity and the Modification Condition

This paper pursues a Vector Space Semantics (VSS) analysis of evaluative and extreme adjectives in absolute and comparative constructions, with a particular emphasis on the licensing of measure phrases (MPs) in these environments. I show that the Modification Condition (Winter 2005), which restricts the distribution of MPs with locative/directional PPs and dimensional adjectives, can be extended to account for MP licensing with evaluative and extreme adjectives as well. Importantly, the non-satisfaction of the Modification Condition is entailed when a set of vectors does not exhaust the range of possible values on a particular scale. This observation thus allows us to link a long-standing generalization that scale exhaustivity and MP licensing are crucially related (Bierwisch 1989) with the formal denotational properties of certain linguistic expressions.

A Semantic Classification Method of Digital Image Based on Cloud Model

In view of randomness and fuzziness of digital image semantics, a new semantic representation of digital image based on cloud model is proposed and a semantic vector space is constructed. In the space, semantic classifications of digital images are completed by calculating the semantic class certainty degree (SCCD). In addition, we propose cloud support vector machine based on image semantics (CSVM-IS) model, which can effectively utilize the knowledge of SCCD. This method can effectively classify the multi-semantic information and eliminate the rejection of the classification samples. Experimental results show that CSVM-IS is superior to the Nesting Algorithm in terms of classification performance.

An Automatic Classification Algorithm of Digital Image Based on Semantics

The randomness and fuzziness of digital image semantics are hot issues in the field of classification technology, so a new semantic representation of digital image based on cloud model is proposed and a semantic vector space is con- structed. In the space, semantic classifications of digital images are completed by calculating the semantic class certainty degree (SCCD). In addition, we propose cloud support vector machine based on image semantics (CSVM-IS) model, which can effectively utilize the knowledge of SCCD. At the same time, CSVM-IS can automatically and effectively clas- sify the multi-semantic information and eliminate the rejection of the classification samples. We design two kinds of simulation experiments to verify the classification performance. One experiment is to verify the classification accuracy, and the other is to verify the classification effectiveness. Experimental results show that CSVM-IS is superior to the Nest- ing Algorithm in terms of classification performance.

A Personalized Product Recommender System based on Semantic Similarity and TOPSIS Method

A personalized product recommendation is an effective mechanism to overcome information overload occurred when customers conduct Internet shopping. The paper presents a personalized recommender system, which integrates semantic similarity computation and TOPSIS method. First, semantic similarity is computed by constructing semantic vector-space, in order to realize the semantic content-based filtering between the product contents and customer profiles. Besides, TOPSIS method is also utilized to construct the comparison mechanism of products by calculating the utility value of each candidate product. Finally, the experiment is conducted to evaluate its recommender quality and the results show the system can give sensible recommendations and is capable of helping customers save enormous time for Internet shopping.

Semantic Vector Models and Functional Models for Pregroup Grammars

We show that vector space semantics and functional semantics in two-sorted first order logic are equivalent for pregroup grammars. We present an algorithm that translates functional expressions to vector expressions and vice-versa. The semantics is compositional, variable free and invariant under change of order or multiplicity. It includes the semantic vector models of Information Retrieval Systems and has an interior logic admitting a comprehension schema. A sentence is true in the interior logic if and only if the ‘usual’ first order formula translating the sentence holds. The examples include negation, universal quantifiers and relative pronouns.

A novel sentence similarity measure for semantic-based expert systems

A novel sentence similarity measure for semantic based expert systems is presented. The well-known problem in the fields of semantic processing, such as QA systems, is to evaluate the semantic similarity between irregular sentences. This paper takes advantage of corpus-based ontology to overcome this problem. A transformed vector space model is introduced in this article. The proposed two-phase algorithm evaluates the semantic similarity for two or more sentences via a semantic vector space. The first phase built part-of-speech (POS) based subspaces by the raw data, and the latter carried out a cosine evaluation and adopted the WordNet ontology to construct the semantic vectors. Unlike other related researches that focused only on short sentences, our algorithm is applicable to short (4–5 words), medium (8–12 words), and even long sentences (over 12 words). The experiment demonstrates that the proposed algorithm has outstanding performance in handling long sentences with complex syntax. The significance of this research lies in the semantic similarity extraction of sentences, with arbitrary structures.

Latent semantic information measurement of corpus orientation

The authors defined an information measurement associated with a topic or semantics for a keyword.Firstly,the topic-based corpus was obtained.Then the latent semantic vector space model of the corpus was established.After that,the information measurement of the keyword was defined through the model.Accordingly,the amount of the topic information any document contained could be calculated.Lastly,the membership measurement which measured the membership degree of the document belonging to the topic was introduced.A measurement threshold was set,thereby it determined whether the documents belonging to the topic or not.The experimental results show that the definition of the information measurement can get over the difficulty of the word-match search and really reach the goal of the semantic-match search.

Bulgarian Spatial Prefixes and Event Structure

In this paper, I explore the combination possibilities of Bulgarian directional prefixes with various motion verbs. Adopting Ramchand’s (in press) event decomposition, Zwarts’ (2005) vector space semantics for directional prepositions, and drawing on various discussions regarding the manner component in the verbal meaning, I propose an analysis that captures the distribution of Goal and Source prefixes. I show how this proposal accounts for the change in the syntactic behavior of prefixed motion verbs compared to their unprefixed counterparts. The proposal also explains the syntactic properties exhibited by verbs when prefixed by different prefixes. I offer a unified treatment of path structure and event structure and suggest that directional prepositions and directional prefixes are semantically identical and originate in the extended PP. The differences between them are due to the syntactic structure in which they participate.

Document clustering using locality preserving indexing and support vector machines

A method of document clustering based on locality preserving indexing (LPI) and support vector machines (SVM) is presented. The document space is generally of high dimensionality, and clustering in such a high-dimensional space is often infeasible due to the curse of dimensionality. In this paper, by using LPI, the documents are projected into a lower-dimension semantic space in which the documents related to the same semantic are close to each other. Then, by using SVM, the vectors in semantic space are mapped by means of a Gaussian kernel to a high-dimensional feature space in which the minimal enclosing sphere is searched. The sphere, when mapped back to semantics space, can separate into several independent components by the support vectors, each enclosing a separate cluster of documents. By combining the LPI and SVM, not only higher clustering accuracies in a more unsupervised effective way, but also better generalization properties can be obtained. Extensive demonstrations are performed on the Reuters-21578 and TDT2 data sets.

A Semantic Service Discovery Network for Large-Scale Ubiquitous Computing Environments

This paper presents an efficient semantic service discovery scheme called UbiSearch for a large-scale ubiquitous computing environment. A semantic service discovery network in the semantic vector space is proposed where services that are semantically close to each other are mapped to nearby positions so that the similar services are registered in a cluster of resolvers. Using this mapping technique, the search space for a query is efficiently confined within a minimized cluster region while maintaining high accuracy in comparison to the centralized scheme. The proposed semantic service discovery network provides a number of novel features to evenly distribute service indexes to the resolvers and reduce the number of resolvers to visit. Our simulation study shows that UbiSearch provides good semantic searchability as compared to the centralized indexing system. At the same time, it supports scalable semantic queries with low communication overhead, balanced load distribution among resolvers for service registration and query processing, and personalized semantic matching.

Non-existential indefinites and semantic incorporation of PP complements

No abstract.

A Landmark-Based Scalable Semantic Resource Discovery Scheme

In this paper, we propose an efficient resource discovery scheme for large-scale ubiquitous computing environments, which supports scalable semantic searches and load balancing among resource discovery resolvers. Here, the resources are described based on the concepts defined in the ontological hierarchy. To semantically search the resources in a scalable manner, we propose a semantic vector space and semantic resource discovery network in which the resources are organized based on their respective semantic distances. Most importantly, landmarks are introduced for the first time to reduce the dimensionality of the vector space. Computer simulation with CAN verifies the effectiveness of the proposed scheme.

정보검색 성능 향상을 위한 단어 중의성 해소 모형에 관한 연구

This paper presents a semantic vector space retrieval model incorporating a word sense disambiguation algorithm in an attempt to improve retrieval effectiveness. Nine Korean homonyms are selected for the sense disambiguation and retrieval experiments. The total of approximately 120,000 news articles comprise the raw test collection and 18 queries including homonyms as query words are used for the retrieval experiments. A Naive Bayes classifier and EM algorithm representing supervised and unsupervised learning algorithms respectively are used for the disambiguation process. The Naive Bayes classifier achieved disambiguation accuracy. while the clustering performance of the EM algorithm is on the average. The retrieval effectiveness of the semantic vector space model incorporating the Naive Bayes classifier showed precision achieving about improvement. However, the retrieval effectiveness of the EM algorithm-based semantic retrieval is lower than the baseline retrieval without disambiguation. It is worth noting that the performances of disambiguation and retrieval depend on the distribution patterns of homonyms to be disambiguated as well as the characteristics of queries.

Network Attack Scenarios Extraction and Categorization by Mining IDS Alert Streams

The past few years have witnessed significant increase in DDoS attacks on Internet, prompting network security as a great concern. With the attacks getting more sophisticated, automatically reasoning the attack scenarios in real time and cat- egorizing those scenarios become a critical challenge. However,the overwhelming flow of events generated by Intrusion Detection System (IDS) sensors make it hard for se- curity administrators to uncover hidden attack plans. This paper presents a semantic vector space model to extract and categorize attack scenarios based on First-order Log- ics (FOL) and linguistics. The modified Case Grammar is introduced to formalize the heterogeneous IDS alerts into uniform structured alert streams. The attack resolution is then used to generate attack semantic network. Afterwards, mutual information is used to determine the alert semantic context range. Based on the attack ontology and alert contexts, attack scenarios are extracted and the alerts are represented as attack semantic space vectors. Finally text categorization technique are used to categorize the intrusion stages. The preliminary results show our model has better performance than the traditional alert correlations.

Mining call center dialog data

We consider the problem of mining conversations between customers and call center representatives for automatically classifying calls into predefined categories. We analyze the conversations for speaker dependent information content using several multi-class classification technologies. The data consists of 539 manually transcribed conversations belonging to 15 categories. Classifiers were built using Support Vector Machines, Naive Bayes, Latent Semantic Analysis, Vector Space and K-Nearest Neighbor technologies. SVM classifiers were found to perform consistently well giving an accuracy of about 74% on the entire data and about 92% when considering only the 4 largest classes. It is observed that very high weightage to either the customer part of the dialog or that of the agent results in poor accuracy. Nearly equal weightage to the customer and agent provides the best results consistently. This approach has potential to identify cross sell and up-sell opportunities in real-time.