Exploiting conversational features to detect high-quality blog comments
read more
Citations
Discovering High-Quality Threaded Discussions in Online Forums
Structure Matters: Adoption of Structured Classification Approach in the Context of Cognitive Presence Classification
Stance Prediction for Russian: Data and Analysis
Automatic Prediction of Comment Quality
References
Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data
Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition
An Introduction to Conditional Random Fields
Related Papers (5)
Sentiment Analysis of Micro-blog Integrated on Explicit Semantic Analysis Method
Finding Opinionated Blogs Using Statistical Classifiers and Lexical Features
Frequently Asked Questions (10)
Q2. What other classes of features were used in this study?
In addition to simple unigram (bag-of-words) features, the authors experimented with two other classes of features: lexical similarity, and conversational features.
Q3. What is the purpose of this work?
Since the goal of this work was to identify high-quality comments, most of their experiments were conducted with comments grouped into GOOD and BAD.
Q4. What is the purpose of the paper?
Future work will focus on refining their ability to classify comments, and incorporating this into an abstractive summarization system.
Q5. What is the main purpose of this paper?
As the amount of content available on the Internet continues to increase exponentially, the need for tools which can analyze and summarize large amounts of text has become increasingly pronounced.
Q6. What is the way to summarise a text?
Several works have shown the effectiveness of CRFs on similar Natural Language Processing tasks which involve sequential dependencies ([1], [4]). [11] uses Linear-Chain CRFs to classify summary sentences to create extractive summaries of news articles, showing their effectiveness on this task. [6] test CRFs against two other classifiers (Support Vector Machines and Naive-Bayes) on the task of classifying dialogue acts in livechat conversations.
Q7. What effect does the classifier have on the unigram features?
since the unigram- and, more notably, similarity-features can still perform quite well without use of the conversational features, their method is not overly-dependent on this effect.
Q8. What is the purpose of the summarizer?
The summarizer then selects a set of messages which maximize a function encompassing information about the sentences in which messages appear, and passes these messages to the NLG system.
Q9. What is the effect of a comment on the conversation tree?
This makes some intuitive sense for training, as comments higher in the conversation tree are likely more important to the conversation as a whole, as the earlier a comment appears in the thread the greater effect it has on the course of conversation down-thread.
Q10. What is the way to evaluate the merged classifications?
Comments which appeared in multiple sequences, and thus received multiple classifications, were marked GOOD if they were classified as GOOD at least once (GOOD if |{ci ∈ C : ci = good}| ≥ 1}, where C is the set of classifications of comment i4.