One of the key challenges in sequential recommendation is how to extract and represent user preferences. Traditional methods rely solely on predicting the next item. But user behavior may be driven by complex preferences. Therefore, these methods cannot make accurate recommendations when the available information user behavior is limited. To explore multiple user preferences, we propose a transformer-based sequential recommendation model, named MrTransformer ( M ulti-p r eference Transformer ). For training MrTransformer, we devise a preference-editing -based self-supervised learning (SSL) mechanism that explores extra supervision signals based on relations with other sequences. The idea is to force the sequential recommendation model to discriminate between common and unique preferences in different sequences of interactions. By doing so, the sequential recommendation model is able to disentangle user preferences into multiple independent preference representations so as to improve user preference extraction and representation. We carry out extensive experiments on five benchmark datasets. MrTransformer with preference editing significantly outperforms state-of-the-art sequential recommendation methods in terms of Recall, MRR, and NDCG. We find that long sequences of interactions from which user preferences are harder to extract and represent benefit most from preference editing.

Click-through rate (CTR) estimation plays as a pivotal function module in various online services. Previous studies mainly apply CTR models to the field of recommendation or online advertisement. Indeed, CTR is also critical in information retrieval, since the CTR probability can serve as a valuable feature for a query-document pair. In this paper, we study the CTR task under account search scenario in WeChat, where users search official accounts or mini programs corresponding to an organization. Despite the large number of CTR models, directly applying them to our task is inappropriate since the account retrieval task has a number of specific characteristics. E.g., different from traditional user-centric CTR models, in our task, CTR prediction is query-centric and does not model user information. In addition, queries and accounts are short texts, and heavily rely on prior knowledge and semantic understanding. These characteristics require us to specially design a CTR model for the task. To this end, we propose a novel CTR prediction model named Knowledge eNhanced hIerarchical Fusion nEtwork (KNIFE). Specifically, to tackle the prior information problem, we mine the knowledge graph of accounts as side information; to enhance the representations of queries, we construct a bipartite graph for queries and accounts. In addition, a hierarchical network structure is proposed to fuse the representations of different information in a fine-grained manner. Finally, the representations of queries and accounts are obtained from this hierarchical network and fed into the CTR model together with other features for prediction. We conduct extensive experiments against 12 existing models across two industrial datasets. Both offline and online A/B test results indicate the effectiveness of KNIFE.

Entity representation plays a central role in building effective entity retrieval models. Recent works propose to learn entity representations based on entity-centric contexts, which achieve SOTA performances on many tasks. However, these methods lead to poor representations for unseen entities since they rely on a multitude of occurrences for each entity to enable accurate representations. To address this issue, we propose to learn enhanced descriptional representations for unseen entities by distilling knowledge from distributional semantics into descriptional embeddings. Specifically, we infer enhanced embeddings for unseen entities based on descriptions by aligning the descriptional embedding space to the distributional embedding space with different granularities, i.e., element-level, batch-level and space-level alignment. Experimental results on four benchmark datasets show that our approach improves the performance over all baseline methods. In particular, our approach can achieve the effectiveness of the teacher model on almost all entities, and maintain such high performance on unseen entities.