Deceptive Answer Prediction with User Preference Graph
Li, Fangtao and Gao, Yang and Zhou, Shuchang and Si, Xiance and Dai, Decheng

Article Structure

Abstract

In Community question answering (QA) sites, malicious users may provide deceptive answers to promote their products or services.

Introduction

Currently, Community QA sites, such as Yahoo!

Related Work

In the past few years, it has become a popular task to mine knowledge from the Community QA sites.

Proposed Features

We first view the deceptive answer prediction as a binary-classification problem.

Deceptive Answer Prediction with User Preference Graph

Besides the textual and contextual features, we also investigate the user relationship for deceptive answer prediction.

Experiments

5.1 Experiment Setting 5.1.1 Dataset Construction

Conclusions and Future Work

In this paper, we discuss the deceptive answer prediction task in Community QA sites.

Topics

question answering

Appears in 5 sentences as: Question Answer (1) question answering (4)
In Deceptive Answer Prediction with User Preference Graph
  1. In Community question answering (QA) sites, malicious users may provide deceptive answers to promote their products or services.
    Page 1, “Abstract”
  2. 3.2.1 Question Answer Relevance The main characteristic of answer in Community
    Page 3, “Proposed Features”
  3. Figure l (a) shows the general process in a question answering
    Page 4, “Deceptive Answer Prediction with User Preference Graph”
  4. Based on the two above assumptions, we can extract three user preference relationships (with same preference) from the question answering example in Figure l (a): m N u5, W N U6, ul N ug, as shown in Figurel (b).
    Page 5, “Deceptive Answer Prediction with User Preference Graph”
  5. Confucius is a community question answering site, developed by Google.
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention question answering.

See all papers in Proc. ACL that mention question answering.

Back to top.

translation model

Appears in 5 sentences as: Translation Model (1) translation model (4)
In Deceptive Answer Prediction with User Preference Graph
  1. Translation Model
    Page 3, “Proposed Features”
  2. A translation model is a mathematical model in which the language translation is modeled in a statistical way.
    Page 3, “Proposed Features”
  3. We train a translation model (Brown et al., 1990; Och and Ney, 2003) using the Community QA data, with the question as the target language, and the corresponding best answer as the source language.
    Page 3, “Proposed Features”
  4. With translation model , we can compute the translation score for new question and answer.
    Page 3, “Proposed Features”
  5. The larger question threads data is employed for feature learning, such as translation model , and topic model training.
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention translation model.

See all papers in Proc. ACL that mention translation model.

Back to top.

Unigrams

Appears in 5 sentences as: unigram (1) Unigrams (2) unigrams (2)
In Deceptive Answer Prediction with User Preference Graph
  1. 3.1.1 Unigrams and Bigrams The most common type of feature for text classi-
    Page 2, “Proposed Features”
  2. feature selection method X2 (Yang and Pedersen, 1997) to select the top 200 unigrams and bigrams as features.
    Page 3, “Proposed Features”
  3. The top ten unigrams related to deceptive answers are shown on Table 1.
    Page 3, “Proposed Features”
  4. Table 1: Top 10 Deceptive Related Unigrams
    Page 3, “Proposed Features”
  5. Besides unigram and bigram, the most effective textual feature is URL.
    Page 7, “Experiments”

See all papers in Proc. ACL 2013 that mention Unigrams.

See all papers in Proc. ACL that mention Unigrams.

Back to top.

topic model

Appears in 4 sentences as: Topic Model (1) topic model (2) topic models (1)
In Deceptive Answer Prediction with User Preference Graph
  1. Topic Model
    Page 3, “Proposed Features”
  2. To reduce the false negatives of word mismatch in vector model, we also use the topic models to extend matching to semantic topic level.
    Page 3, “Proposed Features”
  3. The topic model , such as Latent Dirichlet Allocation (LDA) (Blei et al., 2003), considers a collection of documents with K latent topics, where K is much smaller than the number of words.
    Page 3, “Proposed Features”
  4. The larger question threads data is employed for feature learning, such as translation model, and topic model training.
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention topic model.

See all papers in Proc. ACL that mention topic model.

Back to top.

Bigrams

Appears in 3 sentences as: bigram (1) Bigrams (1) bigrams (1)
In Deceptive Answer Prediction with User Preference Graph
  1. 3.1.1 Unigrams and Bigrams The most common type of feature for text classi-
    Page 2, “Proposed Features”
  2. feature selection method X2 (Yang and Pedersen, 1997) to select the top 200 unigrams and bigrams as features.
    Page 3, “Proposed Features”
  3. Besides unigram and bigram , the most effective textual feature is URL.
    Page 7, “Experiments”

See all papers in Proc. ACL 2013 that mention Bigrams.

See all papers in Proc. ACL that mention Bigrams.

Back to top.

loss function

Appears in 3 sentences as: loss function (2) loss functions (1)
In Deceptive Answer Prediction with User Preference Graph
  1. where L(waZ-,yi) is a loss function that measures discrepancy between the predicted label wT - x,- and the true label yi, where yz- 6 {+1, —l}.
    Page 6, “Deceptive Answer Prediction with User Preference Graph”
  2. The common used loss functions include L(p, y) = (p — y)2 (least square), L(p, y) = ln (1 + exp (—py)) (logistic regression).
    Page 6, “Deceptive Answer Prediction with User Preference Graph”
  3. For simplicity, here we use the least square loss function .
    Page 6, “Deceptive Answer Prediction with User Preference Graph”

See all papers in Proc. ACL 2013 that mention loss function.

See all papers in Proc. ACL that mention loss function.

Back to top.

objective function

Appears in 3 sentences as: objective function (2) objective function: (1)
In Deceptive Answer Prediction with User Preference Graph
  1. The best parameters w* can be found by minimizing the following objective function:
    Page 6, “Deceptive Answer Prediction with User Preference Graph”
  2. The new objective function is changed as:
    Page 6, “Deceptive Answer Prediction with User Preference Graph”
  3. In the above objective function , we impose a user graph regularization term
    Page 6, “Deceptive Answer Prediction with User Preference Graph”

See all papers in Proc. ACL 2013 that mention objective function.

See all papers in Proc. ACL that mention objective function.

Back to top.

PageRank

Appears in 3 sentences as: PageRank (3) PageRank: (1)
In Deceptive Answer Prediction with User Preference Graph
  1. PageRank Score: We employ the PageRank (Page et al., 1999) score of each URL as popularity score.
    Page 3, “Proposed Features”
  2. We compute the user’s authority score (AS) based on the link analysis algorithm PageRank:
    Page 4, “Proposed Features”
  3. When using the user graph as feature, we compute the authority score for each user with PageRank as shown in Equation 1.
    Page 8, “Experiments”

See all papers in Proc. ACL 2013 that mention PageRank.

See all papers in Proc. ACL that mention PageRank.

Back to top.

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