Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
Wang, Baoxun and Wang, Xiaolong and Sun, Chengjie and Liu, Bingquan and Sun, Lin

Article Structure

Abstract

Quantifying the semantic relevance between questions and their candidate answers is essential to answer detection in social media corpora.

Introduction

In natural language processing (NLP) and information retrieval (IR) fields, question answering (QA) problem has attracted much attention over the past few years.

Related Work

The value of the naturally generated question-answer pairs has not been recognized until recent years.

The Deep Belief Network for QA pairs

Due to the feature sparsity and the low word frequency of the social media corpus, it is difficult to model the semantic relevance between questions and answers using only co-occurrence features.

Learning with Homogenous Data

In this section, we propose our strategy to make our DBN model to detect answers in both cQA and forum datasets, while the existing studies focus on one single dataset.

Experiments

To evaluate our question-answer semantic relevance computing method, we compare our approach with the popular methods on the answer detecting task.

Conclusions

In this paper, we have proposed a deep belief network based approach to model the semantic relevance for the question answering pairs in social community corpora.

Topics

social media

Appears in 11 sentences as: social media (11)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. Quantifying the semantic relevance between questions and their candidate answers is essential to answer detection in social media corpora.
    Page 1, “Abstract”
  2. Obviously, these natural QA pairs are usually created during people’s communication via Internet social media , among which we are interested in the community-driven
    Page 1, “Introduction”
  3. In this paper, a novel approach for modeling the semantic relevance for QA pairs in the social media sites is proposed.
    Page 1, “Introduction”
  4. As mentioned above, the user generated questions and their answers via social media are always short texts.
    Page 1, “Introduction”
  5. The semantic information learned from cQA corpus is helpful to detect answers in forums, which makes our model show good performance on social media corpora.
    Page 2, “Introduction”
  6. (2009) both propose the strategies to detect questions in the social media corpus, which is proved to be a nontrivial task.
    Page 3, “Related Work”
  7. Due to the feature sparsity and the low word frequency of the social media corpus, it is difficult to model the semantic relevance between questions and answers using only co-occurrence features.
    Page 3, “The Deep Belief Network for QA pairs”
  8. In the social media corpora, the answers are always descriptive, containing one or several sentences.
    Page 3, “The Deep Belief Network for QA pairs”
  9. Our motivation of finding the homogenous question-answer corpora from different kind of social media is to guarantee the model’s performance and avoid hand-annotating work.
    Page 4, “Learning with Homogenous Data”
  10. It indicates that the word distributions of the two corpora are quite similar, although they come from different social media sites.
    Page 5, “Learning with Homogenous Data”
  11. The task of detecting answers in social media corpora suffers from the problem of feature sparsity seriously.
    Page 6, “Learning with Homogenous Data”

See all papers in Proc. ACL 2010 that mention social media.

See all papers in Proc. ACL that mention social media.

Back to top.

Cosine Similarity

Appears in 8 sentences as: Cosine Similarity (6) Cosine similarity (1) cosine similarity (1)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. Because of this situation, the traditional relevance computing methods based on word co-occurrence, such as Cosine similarity and KL—divergence, are not effective for question-
    Page 1, “Introduction”
  2. Cosine Similarity .
    Page 6, “Experiments”
  3. Given a question q and its :andidate answer 3, their cosine similarity can be :omputed as follows:
    Page 6, “Experiments”
  4. Method P@1(%) MRR (%) Nearest Answer 21.25 38.72 Cosine Similarity 23.15 43.50 HowNet 22.55 41.63 KL divergence 25 .30 51.40 DBN (without FT) 41.45 59.64 DBN (with FT) 45.00 62.03
    Page 7, “Experiments”
  5. Because the features for QA pairs are quite sparse and the content words in the questions are usually morphologically different from the ones with the same meaning in the answers, the Cosine Similarity method become less powerful.
    Page 7, “Experiments”
  6. KL-divergence suffers from the same problems with the Cosine Similarity method.
    Page 7, “Experiments”
  7. Compared with the Cosine Similarity method, this approach has achieved the improvement of 9.3% in P@1, but it performs much better than the other baseline methods in MRR.
    Page 7, “Experiments”
  8. Method P@1 (%) MRR (%) Nearest Answer 36.05 56.33 Cosine Similarity 44.05 62.84 HowNet 41.10 58.75 KL divergence 43.75 63.10 DBN (without FT) 56.20 70.56 DBN (with FT) 58.15 72.74
    Page 8, “Experiments”

See all papers in Proc. ACL 2010 that mention Cosine Similarity.

See all papers in Proc. ACL that mention Cosine Similarity.

Back to top.

content words

Appears in 6 sentences as: content words (6)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. Figure 4 shows the percentage of the concurrent words in the top-ranked content words with high frequency.
    Page 5, “Learning with Homogenous Data”
  2. The number k on the horizontal axis in Figure 4 represents the top k content words in the
    Page 5, “Learning with Homogenous Data”
  3. Percentage of concurrent content words
    Page 5, “Learning with Homogenous Data”
  4. Top k content words
    Page 5, “Learning with Homogenous Data”
  5. Figure 4: Distribution of concurrent content words
    Page 5, “Learning with Homogenous Data”
  6. Because the features for QA pairs are quite sparse and the content words in the questions are usually morphologically different from the ones with the same meaning in the answers, the Cosine Similarity method become less powerful.
    Page 7, “Experiments”

See all papers in Proc. ACL 2010 that mention content words.

See all papers in Proc. ACL that mention content words.

Back to top.

semantic relationship

Appears in 6 sentences as: semantic relationship (5) semantically related (1)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. How to model the semantic relationship between two short texts using simple textual features?
    Page 1, “Introduction”
  2. The network establishes the semantic relationship for QA pairs by minimizing the answer-to-question reconstructing error.
    Page 2, “Introduction”
  3. Judging whether a candidate answer is semantically related to the question in the cQA page automatically is a challenging task.
    Page 2, “Related Work”
  4. The SMT based methods are effective on modeling the semantic relationship between questions and answers and expending users’ queries in answer retrieval (Riezler et al., 2007; Berger et al.,
    Page 2, “Related Work”
  5. In this section, we propose a deep belief network for modeling the semantic relationship between questions and their answers.
    Page 3, “The Deep Belief Network for QA pairs”
  6. The main reason for the improvements is that the DBN based approach is able to learn semantic relationship between the words in QA pairs from the training set.
    Page 7, “Experiments”

See all papers in Proc. ACL 2010 that mention semantic relationship.

See all papers in Proc. ACL that mention semantic relationship.

Back to top.

feature vectors

Appears in 5 sentences as: (2) feature vectors (4)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. In the bottom layer, the binary feature vectors based on the statistics of the word occurrence in the answers are used to compute the “hidden features” in the
    Page 3, “The Deep Belief Network for QA pairs”
  2. j where 0'(x) = 1/ (l + e"‘), 3 denotes the visible feature vector of the answer, qi is the ith element of the question vector, and h stands for the hidden feature vector for reconstructing the questions.
    Page 4, “The Deep Belief Network for QA pairs”
  3. To detect the best answer to a given question, we just have to send the vectors of the question and its candidate answers into the input units of the network and perform a level-by-level calculation to obtain the corresponding feature vectors .
    Page 4, “The Deep Belief Network for QA pairs”
  4. High-dimensional feature vectors with only several nonzero dimensions bring large time consumption to our model.
    Page 6, “Learning with Homogenous Data”
  5. Thus it is necessary to reduce the dimension of the feature vectors .
    Page 6, “Learning with Homogenous Data”

See all papers in Proc. ACL 2010 that mention feature vectors.

See all papers in Proc. ACL that mention feature vectors.

Back to top.

structural features

Appears in 3 sentences as: structural features (3)
In Modeling Semantic Relevance for Question-Answer Pairs in Web Social Communities
  1. Most researchers try to introduce structural features or users’ behavior to improve the models performance, by contrast, the effect of textual features is not obvious.
    Page 2, “Introduction”
  2. The structural features (e. g., authorship, acknowledgement, post position, etc), also called non-textual features, play an important role in answer extraction.
    Page 3, “Related Work”
  3. (2009) show that the structural features have even more contribution than the textual features.
    Page 3, “Related Work”

See all papers in Proc. ACL 2010 that mention structural features.

See all papers in Proc. ACL that mention structural features.

Back to top.

PDF available from www.aclweb.org.
Send us an email