Taking movie film review as an example, using neo4j to analyze the emotion of film commentary in depth study

"Editor's note" with the development of the Internet, users from the previous "read" Web page into the "write" page, the Internet has generated a large number of users to participate in, for such as people, events, products and other valuable comments, and as the network commented on the growth of information explosion, It is difficult to deal with the collection and processing of the massive comment information by the manual method, and the emotion analysis technology comes into being, this article takes the movie film review as an example, uses neo4j to carry on the deep study emotion analysis to the movie commentary.

A movie review site allows users to submit comments about whether they like a movie or not. Fully exploiting these comments and then generating valuable metadata (for relevant content) will provide us with a rare opportunity to understand the user's feelings about the movie in a popular way, which is a cool thing. We can make an objective analysis of subjective content so that we can better understand the trend of products and services, can make better decisions for consumers.

The data model of affective analysis

The main obstacles to achieving these are our structure and transformation data. Current state-of-the-art technologies include naive Bayes, Support Vector Rogue, and Maximum Entropy. The challenge of implementing these technologies remains how to extract features and structured data from text at minimal performance cost, which is what I've decided to focus on.

I use the Feature selection algorithm (click here for details), using graphics database neo4j to solve the challenge of data conversion and usability, while the most advanced natural language parsing algorithm focuses on sentence structure, I decided to design a statistical method for natural language grammar induction, It focuses on the generalization of a large text corpus, generates new features, and uses depth learning to predict the highest probability of a new feature on the left or right side of the current feature.

A graph based NLP instance

I assume that the phrase "one of the worst" has been extracted as a feature of a set of text. The reason for this word extraction is that the phrase has the greatest statistical relevance, which means that the phrase has the best chance of matching after the parent phrase. Using neo4j we can determine the inheritance attribute that produces this word.

Starting at the root node, it is added "{0} {1}", where "one of the worst" will be resolved to (the)-> (ofthe)-> (one of the->).

Such hierarchies will expand to more possibilities, as shown in the following illustration:

In less than a second, this feature selection algorithm can choose the most relevant characteristics and phrases of extraction probability from the text corpus. The reason this technique is very important for affective analysis is that these pattern nodes can connect to their trained text labels, as shown below.

The result of this algorithm is that any natural language text can be parsed at the second level, generating a child graph that can be used for any classification algorithm. This is largely due to the neo4j graph traversal.

Open Source Demo

For the film review example, I selected 500 reviews, including positive and negative labels, using graphify to train a natural language parsing model. In my next blog post, I'm going to show you how to do a better job of classifying film reviews than humans, with a human classification error rate of 30%.

If you want to see fast, click graphify sentiment Analysis for Movie reviews.

Original link: Deep Learning sentiment analysis for Movie reviews using neo4j (Compile/Wei revisers/Zhonghao)