| dc.description.abstract | The rapid growth of conversational recommendation systems has revolutionizedhow users interact with recommender systems. However, accurately capturingand understanding user intent in dynamic conversational settings remains a significantchallenge. This thesis addresses the problem of user intent modeling in aconversational recommendation by proposing a dynamic approach that adapts toevolving user preferences and context.The first contribution of this research is developing a dynamic intent modelingframework using Long Short-Term Memory (LSTM) incorporates both explicitand implicit user signals. By leveraging deep learning algorithms, the proposedframework extracts user intent from conversational data, considering explicit userrequests, implicit preferences, and contextual cues.This thesis introduces a context-driven intent update mechanism to enable dynamicintent modeling. The proposed mechanism updates user intent representationsin real time by continuously monitoring user interactions, contextual factors,and item recommendations. This dynamic modeling approach allows the systemto adapt and refine user preferences as conversations progress, enhancing the accuracyof subsequent recommendations.The proposed dynamic intent modeling framework is evaluated through extensiveexperiments on real-world conversational recommendation datasets. The experimentalresults demonstrate that the dynamic modeling approach significantlyimproves recommendation accuracy compared to traditional static intent modelingmethods. Moreover, the Conversational recommendation approach outperformsbaseline methods, confirming the effectiveness of integrating intent modelingwith LSTM.This thesis explores user intent modeling in recommendation systems by comparingnavigation by preference heuristic algorithm with a proposed LSTM algorithm.The study demonstrates that the LSTM algorithm outperforms navigationby preference, providing more accurate recommendations and higher user satisfaction.The LSTM algorithm leverages sequential modeling to capture temporaldependencies in user-item interactions, resulting in a comprehensive understandingof user intent. The experimental results and user feedback validate the superiorityof the LSTM-based approach, emphasizing its potential for improvingrecommendation accuracy and personalization in recommendation systems. | |
