Word Representation Learning Techniques For Sentiment Analysis

Sentiment analysis is an important research field in natural language processing.In all fields of natural language processing,word representation learning technology is the key research cornerstone.Word representations are a way to represent words in a computer.It can interpret words as low-dimensional vectors.These vectors capture the grammatical and semantic features of words,thus helping the computer understand text.However,existing word vector learning models are trained in the open domain,they are less efficient in professional domains because these word representations lack professional knowledge.They cannot reflect the semantic information it should have in the professional domains.Especially for sentiment analysis,the lack of professional semantics will seriously affect the understanding of text.To solve this problem,this paper uses multi-source knowledge to enrich the semantics of word representations and improve their professionalism.This paper conducts study about word representation learning techniques for sentiment analysis,the specific work contents are as follows:(1)This paper proposes a representation learning framework that utilizes multi-source knowledge fusion to enhance the sentiment semantics of word representations.This framework solves the lack of professional knowledge in sentiment analysis.Firstly,this paper uses multi-source knowledge such as emotional dictionary,syntactic dependency analysis tree,coarse grained text tags to generate sentiment knowledge graphs without manual intervention.These knowledge graphs are rich in statistical information because multiple statistical features are appended to each knowledge item.Based on these statistical features,this article designs an effective knowledge filtering strategy via using Singular Value Decomposition(SVD)algorithm to control the fusion process of emotional knowledge.Moreover,a series of experimental results have shown that compared to the most advanced methods available,the word representations generated by the model proposed in this article can achieve better results in multiple sentiment analysis tasks.(2)Aiming at the real-time and variability of sentiment knowledge,this paper designs and implements a word representation application system for sentiment analysis.The system includes domain knowledge building module,knowledge persistence module and multisource knowledge decision-making module.The domain knowledge building module subscribes to the user’s text information on the social platform by using the Message Queuing Telemetry Transport(MQTT)protocol,and then performs knowledge cleaning and statistical information extraction.The knowledge persistence module performs knowledge persistence and disaster tolerance processing and provides external query interface.The multi-source knowledge decision-making module integrates multi-source knowledge from the Internet and provides conceptual guidance for users,and then relies on the knowledge persistence module to provide real-time sentiment knowledge for users.Each module of the system has high cohesion and low coupling between modules.So as to enhance the independence of the module and the scalability of the system.The system can realize real-time updates after the completion of knowledge base construction,and maintain the effectiveness and real-time of knowledge.