• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
Student
Title
Supervisor
Faculty
Educational Programme
Final Grade
Year of Graduation
German Sokolov
Deep Learning for Aspect-Based Sentiment Analysis
Data Science
(Master’s programme)
9
2017
In this paper, multiple popular deep learning algorithms are investigated with regard to the problem of the aspect analysis of reviews using data from the SentiRuEval contest. Among them are convolutional (CNN) and recurrent neural networks (RNN).

Based on the results of the extraction of aspect terms, it was found that some neural network architectures can significantly overperform widely used CRF method. At the same time, neural networks have a significant advantage in training, because they do not require the manual generation of multiple feautres and are able to adapt for use in other subject areas and natural languages.

However, in the problem of determining the tonality of an aspect term or an aspect group, deep learning algorithms significantly underperform the results of other more traditional classifiers - SVM and Gradient Boosting Classifier. Convolutional neural networks showed a much worse result than RNN.

In almost all cases the most optimal architecture based on F1-measure was LSTM recurrent neural network, in other rare situations - GRU. For the task of extracting aspects, the configuration with one layer of hidden state and direct connections was more effective. For the task of sentiment analysis, it was preferable to use bidirectional recurrent neural networks.

Experiments have confirmed the hypothesis that the use of a specialized collection of texts on a given topic has a significant positive effect on the quality of the algorithm. In addition, the use of POS-tags has a significant positive effect in solving all tasks of the competition. Additional consideration of the symbols of emotions can also be useful in analizing sentiment.

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses