An Application Of Word Embeddings In Recommending Alternative Query Terms In Domain-Speci c Search

Keywords
Loading...
Thumbnail Image
Issue Date
2019-06-01
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A Query is a statement of a requester specifying their information need. A poorly formed query can be ambiguous, which can lead to poor performance of an information retrieval machine. Aiding the user by sug- gesting di erent query terms could be of use in avoiding this problem. A common way of nding query recommendations is by using query logs (Baeza-Yates et al.). However, smaller companies and institutions that operate in a speci c domain rarely possess such query logs as they require large user-bases. Instead of using logs, one could use a language model to nd semantically similar terms to the input. A popular example of such a model is a word embeddings (Word2Vec, Mikolov et al.) model. This technique uses a neural network to encode word features to real vectors based on neighboring words in texts of a corpus. These vectors can be compared, so similar words to an input can be extracted. This research proposes a system that can recommend words based on single query terms provided by a user. This system functions as an add-on to an existing domain-speci c search engine. A model was trained as part of this thesis and its quality was evaluated. Furthermore, the model was used in a rec- ommendation system and subsequently experimented with. No signi cant evidence was found regarding a performance gain in this thesis. Improve- ments are proposed that could potentially lead to a signi cant result in the future.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen