In April, the world is full of fragrance, and Fang Fei is full @_@
In order to have a more complete understanding of sentiment analysis methods, this article is mainly divided into two parts:
1. Stanford University Natural Language Processing Lesson 7 "Sentiment Analysis" Click to open the link
Second, the latest sentiment analysis related papers: deep context, support vector machine, two-level LSTM, multimodal sentiment analysis, software engineering, code mixingClick to open the link
-----------------------------------------------------------------------------------------------------------------------------------------------------
1. Sentiment Analysis, Course 7 of Natural Language Processing at Stanford University
The sentiment dictionary and the method of constructing the sentiment dictionary are recorded here. After all, the dictionary-based method is recommended when the amount of data is not large.
Among them, the dictionary construction methods specifically include: bootstrapping-based methods, Turney Algorithm, Using WordNet, etc.
The dictionary approach is more primitive, but not entirely without reference value.
Second, the latest sentiment analysis related papers: deep context, support vector machine, two-level LSTM, multimodal sentiment analysis, software engineering, code mixing
The sentiment analysis method using deep learning has not been introduced before. Although sentiment analysis is mostly attributed to the problem of sentiment classification, the method of deep learning is novel and worth studying.
1. "ρ-hot Lexicon Embedding-based Two-level LSTM for Sentiment Analysis" ( two-level LSTM sentiment analysis based on ρ-hot dictionary Emebedding ), using a new labeling strategy and two-layer LSTM to build a sentiment classifier. First, a two-stage labeling strategy is introduced for sentiment texts. In the first stage, annotators are invited to annotate large volumes of short texts with relatively pure emotional orientation. Each sample is marked by only one annotator. In the second stage, a relatively small number of text samples with mixed emotion orientations are annotated, each sample is marked by multiple annotators. Second, a two-level Long Short-Term Memory (LSTM) network is proposed to achieve two-level feature representation and classify the sentiment orientation of text samples to utilize two labeled datasets. Finally, in the proposed two-stage LSTM network, lexical embeddings are utilized to incorporate linguistic features used in dictionary-based methods.
Two-layer LSTM network:
2. Multimodal Sentiment Analysis: Addressing Key Issues and Setting up Baselines
