Multi-Point Semantic Representation for Intent Classification

Zhang, Jinghan; Ye, Yuxiao; Zhang, Yue; Qiu, Likun; Fu, Bin; Li, Yang; Yang, Zhenglu; Sun, Jian

doi:10.1609/aaai.v34i05.6498

Cited by 21 publications

(34 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Several works tried to exploit an external knowledge base to accomplish dataless classification (Chang et al, 2008;Song and Roth, 2014). However, external knowledge base like Wikipedia is not always available for many languages or domains.…”

Section: Related Workmentioning

confidence: 99%

“…Recent studies on dataless text classification show promising results on reducing labeling effort (Liu et al, 2004;Druck et al, 2008;Chang et al, 2008;Hingmire et al, 2013;Hingmire and Chakraborti, 2014;Song and Roth, 2014;Chen et al, 2015;Li et al, 2016b;Li et al, 2018;Li et al, 2019a;Shalaby and Zadrozny, 2019). Without any labeled documents, a dataless classifier performs text classification by using a small set of relevant words for each category (called "seed words") or resorting to hidden topic labeling.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Filtering and Classifying Relevant Short Text with a Few Seed Words

Chen

Yan

2019

Data and Information Management

View full text Add to dashboard Cite

Filtering out irrelevant documents and classifying the relevant ones into topical categories is a de facto task in many applications. However, supervised learning solutions require extravagant human efforts on document labeling. In this paper, we propose a novel seed-guided topic model for dataless short text classification and filtering, named SSCF. Without using any labeled documents, SSCF takes a few “seed words” for each category of interest, and conducts short text filtering and classification in a weakly supervised manner. To overcome the issues of data sparsity and imbalance, the short text collection is mapped to a collection of pseudodocuments, one for each word. SSCF infers two kinds of topics on pseudo-documents: category-topics and general-topics. Each category-topic is associated with one category of interest, covering the meaning of the latter. In SSCF, we devise a novel word relevance estimation process based on the seed words, for hidden topic inference. The dominating topic of a short text is identified through post inference and then used for filtering and classification. On two real-world datasets in two languages, experimental results show that our proposed SSCF consistently achieves better classification accuracy than state-of-the-art baselines. We also observe that SSCF can even achieve superior performance than the supervised classifiers supervised latent dirichlet allocation (sLDA) and support vector machine (SVM) on some testing tasks.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Filtering and Classifying Relevant Short Text with a Few Seed Words

Chen

Yan

2019

Data and Information Management

View full text Add to dashboard Cite

show abstract

“…Entity-based text representation may be utilized in many other tasks, e.g., computing document similarity [44], text classification [10], or question answering [6,45]. Medical search is another prominent example for the use of controlled vocabulary representations, with a lot of work conducted in the context of the TREC Genomics track [31,36,46].…”

Section: Further Readingmentioning

confidence: 99%

Leveraging Entities in Document Retrieval

Balog

2018

Entity-Oriented Search

View full text Add to dashboard Cite

This chapter focuses on the classic IR task of ad hoc document retrieval and discusses how entities may be leveraged to improve retrieval performance. At their core, all document retrieval methods compare query and document representations. Traditionally, these representations are based on terms (words). Entities facilitate a semantic understanding of both the user's information need, as expressed by the keyword query, and of the document's content. Entities thus may be used to improve query and/or document representations. As a first step of that process, entities that are related to the query need to be identified, thereby establishing a mapping between queries and entities. As we shall explain in Sect. 8.1, one may go beyond considering only those entities that are explicitly mentioned in the query. We next present three different families of approaches, which are illustrated in Fig. 8.1. • Expansion-based methods utilize entities as a source of expansion terms to enrich the representation of the query (Sect. 8.2). • Projection-based methods treat entities as a latent layer, while leaving the original document/query representations intact (Sect. 8.3). • Entity-based methods consider explicitly the entities that are recognized in documents, as first-class citizens, and embrace entity-based representations in "duet" with traditional term-based representations in the retrieval model (Sect. 8.4). This particular order corresponds to the temporal evolution of research in this area, where the tendency toward more and more explicit entity semantics is clearly reflected. Throughout this chapter, we shall assume that both queries and documents have been annotated with entities, using entity linking techniques we have discussed before (see Chap. 5 for documents and Sect. 7.3 for queries). A particular challenge involved here is how to deal with the uncertainty of these automatic annotations. In practice, necessitated by efficiency considerations, all methods described in this chapter are implemented as re-ranking mechanisms. The details are found in Sect. 8.5. Finally, we present standard datasets and useful resources in Sect. 8.6. We refer to Table 8.1 for the notation used throughout this chapter.

show abstract

“…Uma das possíveis aplicações para o ESA (e o ESA-G) é a classificação de textos [49]. No domínio em estudo, um texto pode se relacionar, simultaneamente, a vários diferentes rótulos, caracterizando a então denominada "Classificação Multirrótulo" apresentada a seguir.…”

Section: Esa-gunclassified

“…A proposta inicial do ESA não tem um classificador, dado que seu propósito original não é a categorização de documentos e sim a medição de similaridade semântica entre eles. Em [49], apresenta-se uma das primeiras propostas de classificação com o ESA sem um conjunto de treinamento explícito, baseando-se apenas no conhecimento incorporado em seu sistema de conceitos para realizar categorizações. Como o problema apresentado na Seção 1.1 envolve a análise de fragmentos de texto e não a simples comparação entre dois ou mais trechos de documentos, uma das primeiras mudanças introduzidas no ESA foi a de retornar, no lugar do cosseno entre dois textos, o conceito que mais se aproxima semanticamente do texto submetido ao A2E.…”

Section: Adaptação No Classificador Do Esaunclassified

Avaliação semântica da integração da gestão de riscos de segurança em documentos de software da administração pública

Peclat¹

View full text Add to dashboard Cite

Brasília 2015 DedicatóriaEsse trabalho é primeiramente dedicado à Deus, que me permitiu chegar ao final dele, quando muitas vezes, duvidei da possibilidade de fazê-lo. Em segundo lugar, é dedicado à minha família, especialmente à minha esposa, Giuliana, que foi obrigada a sofrer com minha "ausência" por conta deste mestrado durante quase dois anos.Aos chefes da Controladoria-Geral da União que acreditaram, desde o início, na importância deste trabalho, ele também é dedicado a vocês.Por último, mas não menos importante, dedico este trabalho ao meu orientador -Dr. Guilherme Ramos -pela compreensão e apoio fundamentais em todas as dificuldades enfrentadas.Esse trabalho é para vocês. Esse trabalho é por vocês.iv AgradecimentosEm primeiro lugar, agradeço à Controladoria-Geral da União por permitir minha participação nesse programa de mestrado. Em seguida, um sincero agradecimento ao corpo docente da Universidade de Brasília que proporcionou, ao longo deste curso, uma extensão da minha maneira de pensar sobre engenharia de software e seus desafios contemporâneos.Agradecimento especial aos participantes do survey realizado ao longo desta pesquisa, bem como ao e Gestão pelo suporte oferecido.Por último, agredecimentos também são dirigidos a organizações como OWASP, MITRE, SOFTEX e Fatto Consultoria que permitiram, por meio de materiais disponibilizados online, a realização deste trabalho.v Palavras-chave: riscos de segurança, linguagem natural, mineração de texto vi Abstract Secure software has its security risks well managed regarding vulnerabilities. IT managers in the Brazilian Federal Public Sector (APF) are legally required to strive for software security. However, only a small number of software security professionals are employed in the technical support for development, maintenance and acquisition of such software in their public organizations. This problem, combined with recent advances in text mining, especially in natural language processing and multi-label classification, motivate this work on research for a computational solution that can understand the semantics of sentences in documents written in Portuguese and connect them to previously defined software security risks, such as OWASP Top Ten.This solution (A2E) can improve the software factories bidding process of the APF by providing the authors and reviewers of technical specifications with a computational tool which can automatically evaluate the integration between security risks management and software processes described in these documents. After applying A2E to more than 120 thousand sentences extracted from OWASP ASVS and past APF specifications, a survey was conducted to compare its performance with the opinion of software engineers and security specialists through objective metrics like Precision, Recall, Hamming Loss and Negative Predictive Values (NPV).A2E's final version, after a series of improvements in the development process, obtained a significantly better Hamming Loss measure when compared to the specialists' assessments. Additionally, ...

show abstract

Multi-Point Semantic Representation for Intent Classification

Cited by 21 publications

References 9 publications

Filtering and Classifying Relevant Short Text with a Few Seed Words

Filtering and Classifying Relevant Short Text with a Few Seed Words

Leveraging Entities in Document Retrieval

Avaliação semântica da integração da gestão de riscos de segurança em documentos de software da administração pública

Contact Info

Product

Resources

About