ระบบค้นคืนสารสนเทศแบบจัดลำดับและแบบค้นคืนย้อนกลับบนโครงสร้างแถวลำดับแพ็ต / มานพ จงเจริญใจ = An information retrieval system using ranking and relevance feedback on the PAT array / Manoop Jongcharoenjai
This thesis presents a development of information retrieval system using ranking and relevance feedback on PAT arrays which are used as index for retrieval. A PAT array is a structure that fits for Thai text which Thai text which is not completely segmented. PAT arrays store index in semi-infinite strings (sistrings). The development consists of 3 subsystems: the index term weights creation subsystem, the ranking subsystem, and the relevance feedback subsystem. The index term weights creation subsystem stores unique sistrings and frequency of each sistring from all documents in a PAT array in order to decrease retrieval time. The retrieval compares between user’s query and words from sistrings which follow the linguistics rules. After getting the retrieval results, the ranking subsystem calculates term weights for each document, and sort the documents in descending order using the term weights as a key. The relevance feedback subsystem allows the user to select relevant documents, and enter new query in order to improve the results. This research experienced five term weight formulas. The experiments showed that there are two formulas that give the best results. One of the two formulas consists of term frequency in its formula and the other consists of term frequency multiplied by inverse document frequency. For the relevance feedback, the experiments showed that choosing the appropriate threshold help the system promote new queries that help improve better results.
