Search

Abstract Online group buying is a burgeoning business model of Internet shopping, in which people with the same merchandise interests form a group and co-purchase goods with favorable prices. The buyer who launches the co-purchase is called the initiator, and other buyers are called the co-purchasers. Although recommending co-purchasers for a target buyer (co-purchase initiator) on the group buying is an interesting problem, existing studies have paid few attention to this topic. Different from the collaborator recommendation that only considers users with high similarity to the target user, co-purchaser recommendation takes both users with high and weak similarity into account, and the recommendation results can achieve high recall and diversity. However, the task turns out to be a challenging problem since it is hard to make a precise recommendation for buyers with weak similarity. To address the problem, we propose the following two methods. In the first one, we directly impose a penalty to the weak similar co-purchasers in the embedding space. To further improve the recommendation performance, in the second one, we smoothly increase the co-occurrence probability of the weak similar co-purchasers by truncated bias walk. Our experimental results on real datasets show that the proposed methods, particularly the latter, can effectively complete the co-purchaser recommendation and has high recommendation performance. In addition, considering that co-purchase may last longer, the total recommendation result can be generated in multiple stages and adjust the current recommendation list based on the feedback from the recommendation of previous stages. It is a trick for all co-purchaser recommendation methods to make the total result better.

In this work, pulsed ultrasound was used to facilitate steady-state reactive crystallization and increase the final yield and productivity of lithium carbonate in continuously operated single and multistage mixed suspension mixed product removal (MSMPR) crystallizers. Experimental analyses of the stirred tank MSMPR cascade were performed to investigate the effects of ultrasound field, residence time and temperature which contributed to the steady-state yield, crystal size distribution and crystal morphology. The results show that pulsed ultrasound can not only significantly enhance the reaction rate, but also help to improve the particle size distribution and the crystal habit. Subsequently, a population balance model was developed and applied to estimate the final yield of the continuous process of the lithium bicarbonate thermal decomposition reaction coupling lithium carbonate crystallization. The consistency of the final yield between the experiments and the simulations proved the reliability of the established model. Through the experimental and simulation analyses, it is demonstrated that the use of pulsed ultrasound, higher final stage temperature, MSMPR cascade design and appropriate residence time help to achieve higher yield and productivity. Furtherly, based on the conclusion drawn, pulsed ultrasound enhanced three-stage MSMPR cascaded lithium carbonate continuous crystallization processes were designed, and the maximum productivity of 44.0 g/h was obtained experimentally.

Faced with the explosive demand of real-world applications, spatial crowdsourcing has attracted much attention, in which task assignment algorithms take the dominant role in the past few years. On the one hand, most recent studies concentrate on maximizing the overall benefits of the platform, ignoring the fact that user experience also plays an essential role in task allocation. On the other hand, they focus on matching, that is, how to assign tasks, rather than batching, that is, when to make assignment. In fact, user experience also depends on batching, but this is largely overlooked by current studies. In this paper, we propose a self-adaptive batching mechanism to enhance user experience in spatial crowdsourcing. With appropriate start-up timestamps, previous matching methods can perform better. Multi-armed bandit algorithm in reinforcement learning is adopted to split the batch dynamically according to historical current states. Extensive experimental results on both real and synthetic datasets demonstrate the effectiveness and efficiency of the proposed approach.

Point-of-interest (POI) recommendation has attracted many interests recently because of its significant potential for helping users to explore new places and helping location-based service (LBS) providers to carry out precision marketing. Compared with the user-item rating matrix in conventional recommender systems, the user-location check-in matrix in POI recommendation is usually much more sparse, which makes the notorious cold start problem more prominent in POI recommendation. Trustoriented recommendation is an effective way to deal with this problem but it requires that the recommender has access to user check-in and trust data. In practice, however, these data are usually owned by different businesses who are not willing to share their data with the recommender mainly due to privacy and legal concerns. In this paper, we propose a privacy-preserving framework to boost data owners willingness to share their data with untrustworthy businesses. More specifically, we utilize partially homomorphic encryption to design two protocols for privacy-preserving trust-oriented POI recommendation. By offline encryption and parallel computing, these protocols can efficiently protect the private data of every party involved in the recommendation. We prove that the proposed protocols are secure against semi-honest adversaries. Experiments on both synthetic data and real data show that our protocols can achieve privacy-preserving with acceptable computation and communication cost.

Recommender systems, e.g., movie recommendation, play an important role in our life. However, few movie recommendation methods have considered the rich visual content information in posters and still frames, which can be used to alleviate the data sparsity and cold start problems in recommendation. Moreover, no existing paper has taken visual feature learning and recommendation into a unified optimization process. To this end, in this paper, we focus on how to use visual contents to improve the performance of movie recommendation and propose a novel movie recommendation model named unified visual contents matrix factorization (UVMF) that integrates visual feature extraction and recommendation into a unified framework. Specifically, we integrate convolutional neural network into probabilistic matrix factorization, and the model can be trained end-to-end. Moreover, we unfix weights in the last few layers of VGG16 to learn features and adapt them for the movie recommendation task. Finally, the experimental results on real-world data show that UVMF outperforms other benchmark methods in terms of recommendation accuracy.