Deep learning based query suggestion framework for search engines

With the growth in the number of information in the online world and the increase in the number of users, the use of search engines to access information has become increasingly widespread. This increase has encouraged more users to use search engines. With this situation, the query complexity of users has started to increase. It has also encouraged search engines to make and improve query suggestions to improve users' experience. So far, many different methods have been used for the query suggestion task. However, most of the methods used are far from fully understanding users' wishes and the context of the search. Determining target queries is a very difficult issue due to the nature of query suggestions. Capturing the real meanings of user searches is very important to provide accurate and effective query suggestions to users. Transformer architectures perform very well for a wide range of natural language processing tasks and are useful for capturing deep meanings in language-related problems. Transformers are deep neural networks that use the self-attention mechanism to capture contextual relationships within sequential data. In this thesis study, we tried to determine the underlying meanings of the user's searches and the context of the sessions for query suggestion by utilizing the power of transformer-based deep learning architectures. The "transformer-based query suggestion (TQS)" model presented in the study was compared with various historical query suggestion models and analyzes were presented. In addition, the performance results obtained show that the training phase of the proposed model is quite fast.