| Abstract [eng] |
The aim of the project is to create an application for customer review classification in Lithuanian language in the specific E-commerce business domain. Various Natural Language Processing techniques and machine learning algorithms were used for customer reviews evaluation and polarity categorization. Analyzing customer reviews in Lithuanian language makes the task more challenging due to the language specifics. Therefore, additional pre-processing steps were utilized in this work. Integration with external and internal Representational State Transfer Application Programming Interfaces were made in order to create upgradeable classification model and application for data transformation, both integral into an ecosystem based on microservices within any modern organization. From an analytical perspective, classic models such as Support Vector Machine and Random Forests were built and tested with differently pre-processed datasets. In addition to these methods, deep learning models were added for comparison. Best results were achieved using Support Vector Machine algorithm and Deep Neural Network with embedding and flatten layers. Using TensorFlow framework and additional R programming language libraries, the best model was deployed as a microservice, which classifies customer reviews written in Lithuanian in real-time. The final product could be integrated into other applications and services for a variety of text classification use cases. |
