Effective and efficient polarity estimation in blogs based on sentence-level evidence

Abstract

One of the core tasks in Opinion Mining consists of estimating the polarity of the opinionated documents found. In some scenarios (e.g. blogs), this estimation is severely affected by sentences that are off-topic or that simply do not express any opinion. In fact, the key sentiments in a blog post often appear in specific locations of the text. In this paper we propose several effective and robust polarity detection methods based on different sentence features. We show that we can successfully determine the polarity of documents guided by a sentence-level analysis that takes into account topicality and the location in the blog post of the subjective sentences. Our experimental results show that some of our proposed variants are both highly effective and computationally-lightweight.

Multi-lingual support for lexicon-based sentiment analysis guided by semantics

“…However, this task may be approached as a ternary classification problem as well, by introducing a third class of neutral documents. An alternative to such sentiment classification approaches is the determination of a degree of positivity or negativity of natural language text in order to produce, e.g., rankings of positive and negative documents [37,38].…”

Sentiment-Based Ranking of Blog Posts Using Rhetorical Structure Theory

“…We apply an effective and efficient approach [4] based on sentence retrieval and a well-know sentiment classifier, OpinionFinder (OF) [10]. OpinionFinder estimates what sentences are subjective and also marks various aspects of the subjectivity in the sentences, including the source (holder) of the opinions and the words that are included in phrases expressing positive or negative sentiments.…”

“…This second stage can also be subdivided into two different substaks: an opinion-finding task, where the main aim is to find opinionated blog posts related to the query, and a subsequent polarity task to identify the orientation of a blog post with respect to the topic (e.g., positive or negative). For polarity estimation, researchers often apply naive methods (e.g., classifiers based on frequency of positive/negative terms) [4]. Polarity estimation is a really challenging task with many unresolved issues (e.g., irony, conflicting opinions, etc.).…”

“…Therefore, it would not be wise to apply RST on the entire blog posts. We build on recent advances in extracting key opinionated sentences for polarity estimation in blog posts [4] and analyse the structure of the discourse only for selected passages. This is beneficial to avoid noisy chunks of text and it is also convenient from a computational complexity perspective because discourse processing is not lightweight.…”

A weighted lexicon-based generative model for opinion retrieval

“…At the first stage, relevant documents are retrieved by their relevance scores (e.g., TF*IDF ranking or language model). At the second stage, an opinion score (the ranking score of to what extent it is subj ective or obj ective) is calculated for each retrieved document by using machine learning methods, lexicon-based statistical approaches or other methods [5][6][7][8][9][10][11][12]. Finally, an overall score combining the two is computed to re-rank the retrieved documents.…”