Tuesday 08 April 2025
A new approach to document retrieval has been developed, one that combines both the content and headings of a document to provide more accurate search results. This innovative method was tested on a dataset of Thai documents, specifically designed for foreign trade and investment.
The traditional methods of document retrieval often rely solely on the content or headings of a document, which can lead to inaccurate results. For example, a search query might return a document that is relevant but not exactly what the user was looking for. This new approach addresses this issue by considering both the content and headings of a document, allowing for a more comprehensive understanding of the document’s relevance.
The system uses a deep learning model to measure the semantic similarity between a user’s query and a document’s heading. It then combines this information with the BM25 algorithm, which is used to rank documents based on their term frequencies in the content. The result is a final ranking score for each document, which allows the system to return the most relevant results.
The system was tested using a dataset of 406 queries and achieved an accuracy rate of 65.76%. This outperformed both the BM25 method alone (57.88%) and the Word2Vec-based method (62.81%). The user satisfaction survey also showed promising results, with participants rating the system’s usefulness at an average of 3.70 out of 5.
The authors of this study aimed to develop a system that could provide accurate search results for Thai entrepreneurs seeking information on foreign trade and investment. They achieved this by leveraging both the content and headings of documents, allowing for a more comprehensive understanding of the document’s relevance.
This new approach has significant implications for the field of natural language processing and information retrieval. It demonstrates the potential of combining multiple techniques to achieve better results than relying solely on one method. Additionally, it highlights the importance of considering both the content and headings of a document in search queries.
In practical terms, this system could be used by Thai entrepreneurs seeking information on foreign trade and investment. For example, a user might submit a query asking for information on importing goods into Myanmar. The system would then return relevant documents that match their query, providing them with accurate and useful information.
Overall, this new approach to document retrieval has the potential to revolutionize the way we search for information online. By considering both the content and headings of a document, it provides more accurate results than traditional methods.
Cite this article: “ThaiQuery: A Novel Approach to Enhance Information Retrieval in Thai Documents using Deep Learning Techniques”, The Science Archive, 2025.
Document Retrieval, Natural Language Processing, Information Retrieval, Deep Learning Model, Semantic Similarity, Bm25 Algorithm, Term Frequencies, Thai Documents, Foreign Trade And Investment, Search Results.
