Mingwei Zhang

Conversational recommender systems (CRS) facilitate natural language interactions for more effective item suggestions. While these systems show promise, they face challenges in effectively utilizing and integrating informative data with conversation history through semantic fusion. In this study we present an innovative framework for extracting social information from conversational datasets by inferring ratings and constructing user-item interaction and user-user relationship graphs. We introduce a social information sensitive semantic fusion (SISSF) method that employs contrastive learning (CL) to bridge the semantic gap between generated social information and conversation history. We evaluated the framework on two public datasets (ReDial and INSPIRED) using both automatic and human evaluation metrics. Our SISSF framework demonstrated significant improvements over baseline models across all metrics. For the ReDial dataset, SISSF achieved superior performance in recommendation tasks (R@1: 0.062, R@50: 0.437) and conversational quality metrics (Distinct-2: 4.223, Distinct-3: 5.595, Distinct-4: 6.155). Human evaluation showed marked improvement in both fluency (1.81) and informativeness (1.63). We observed similar performance gains on the INSPIRED dataset, with notable improvements in recommendation accuracy (R@1: 0.046, R@10: 0.129, R@50: 0.269) and response diversity (Distinct-2: 2.061, Distinct-3: 4.293, Distinct-4: 6.242). The experimental results consistently validate the effectiveness of our approach in both recommendation and conversational tasks. These findings suggest that incorporating social context through CL can significantly improve the personalization and relevance of recommendations in conversational systems.

Recent advances in interactive recommender systems (IRSs) have sparked great interest in the exploitation of deep reinforcement learning (DRL) techniques to consider long-term user experiences. However, DRL-based IRSs tend to suffer from two common issues. The first is sample efficiency, i.e., the large action space consisting of a large number of candidate items causes that huge amount of interaction data is required to train an effective recommendation policy. The second is lacking explainability of the learned recommendation policy. To this end, we propose a method called Knowledge Graph-enhanced Hierarchical Reinforcement learning for interactive and explainable recommendation (KGHR) to harness the advantages of both DRL and knowledge graphs (KGs) to deal with the above two issues of DRL-based IRSs. Specifically, the proposed model decomposes an interactive recommendation task into a hierarchy of two level reinforcement learning tasks. The high-level agent learns to rank the candidate items sampled by the low-level agent for recommendation step by step, in which a KG is incorporated to enrich the state representation to enhance sample efficiency. The low-level agent guides user-conditional candidate item sampling, in which the same KG is exploited to couple candidate item selection and explainability by providing actual reasoning paths. Comprehensive experiments have been conducted in a simulated online environment with two real-world datasets, which demonstrate the superiority of our approach in terms of accuracy and explainability against state-of-the-art methods.

An increasing number of services are being offered online, which leads to great difficulties in selecting appropriate services during mashup development. There have been many service recommendation studies and achieved remarkable results to alleviate the issue of service selection challenge. However, they are limited to suggesting services only for a single round or the next round, and ignore the interactive nature in real-world service recommendation scenarios. As a result, existing methods can’t capture developers’ shifting requirements and obtain the long-term optimal recommendation performance over the whole recommendation process. In this paper, we propose a deep reinforcement learning based interactive service recommendation model (RLISR) to tackle this problem. Specifically, we formulate interaction service recommendation as a multi-round decision-making process, and design a reinforcement learning framework to enable the interactions between mashup developers and service recommender systems. First, we propose a knowledge-graph-based state representation modeling method, wherein we consider both the positive and negative feedbacks of developers. Then, we design an informative reward function from the perspective of boosting recommendation accuracy and reducing the number of recommendation rounds. Finally, we adopt a cascading Q-networks model to cope with the enormous combinational candidate space and learn an optimal recommendation policy. Extensive experiments conducted on a real-world dataset validate the effectiveness of the proposed approach compared to the state-of-the-art service recommendation approaches.

Personalized trip recommendation attempts to recommend a sequence of Points of Interest (POIs) to a user. Compared with a single POI recommendation, the POIs sequence recommendation is challenging. There are only a couple of studies focusing on POIs sequence recommendations. It is a challenge to generate a reliable sequence of POIs. The two consecutive POIs should not be similar or from the same category. In developing the sequence of POIs, it is necessary to consider the categories of consecutive POIs. The user with no recorded history is also a challenge to address in trip recommendations. Another problem is that recommending the exact and accurate location makes the users bored. Looking at the same kind of POIs, again and again, is sometimes irritating and tedious. To address these issues in recommendation lies in searching for the sequential, relevant, novel, and unexpected (with high satisfaction) Points of Interest (POIs) to plan a personalized trip. To generate sequential POIs, we will consider POI similarity and category differences among consecutive POIs. We will use serendipity in our trip recommendation. To deal with the challenges of discovering and evaluating user satisfaction, we proposed a Serendipity-Oriented Personalized Trip Recommendation (SOTR). A compelling recommendation algorithm should not just prescribe what we are probably going to appreciate but additionally recommend random yet objective elements to assist with keeping an open window to different worlds and discoveries. We evaluated our algorithm using information acquired from a real-life dataset and user travel histories extracted from a Foursquare dataset. It has been observationally confirmed that serendipity impacts and increases user satisfaction and social goals. Based on that, SOTR recommends a trip with high user satisfaction to maximize user experience. We show that our algorithm outperforms various recommendation methods by satisfying user interests in the trip.

With the increasing numbers of Web services and service users on World Wide Web, predicting QoS (Quality of Service) for users will greatly aid service selection and discovery. Due to the different backgrounds and experiences of users, they have different QoS experiences when interacting with the same service. Even two users who have similar experiences on some services can have diverging views when considering services. This paper proposes an approach to predict QoS based on other users' QoS experiences. This method employs similarity mining and prediction from users' experience by firstly selecting a set of web services that have the highest degree of similarity with the target service by comparing the target service with the others services used by target user. Secondly, the missing value can be calculated through the data of similar services. On the basis of that, we calculate the user similarity and predict QoS data for target user. Experimental results show that it can improve the prediction accuracy of QoS for Web service by using this method.