Click-Through Rate (CTR) prediction plays an important role in many industrial applications, such as online advertising and recommender systems. How to capture users' dynamic evolving interests from their behavior sequences remains a continuous research topic the CTR prediction. However, most existing studies overlook intrinsic structure of sequences: are composed sessions, where sessions user behaviors separated by occurring time. We observe that highly homogeneous each session,...

Capturing users' precise preferences is a fundamental problem in large-scale recommender system. Currently, item-based Collaborative Filtering (CF) methods are common matching approaches industry. However, they not effective to model dynamic and evolving of users. In this paper, we propose new sequential deep (SDM) capture by combining short-term sessions long-term behaviors. Compared with existing sequence-aware recommendation methods, tackle the following two inherent problems real-world...

Recommender system, as an essential part of modern e-commerce, consists two fundamental modules, namely Click-Through Rate (CTR) and Conversion (CVR) prediction. While CVR has a direct impact on the purchasing volume, its prediction is well-known challenging due to Sample Selection Bias (SSB) Data Sparsity (DS) issues. Although existing methods, typically built user sequential behavior path "impression->click->purchase", effective for dealing with SSB issue, they still struggle address DS...

Nowadays, the product search service of e-commerce platforms has become a vital shopping channel in people's life. The retrieval phase products determines system's quality and gradually attracts researchers' attention. Retrieving most relevant from large-scale corpus while preserving personalized user characteristics remains an open question. Recent approaches this domain have mainly focused on embedding-based (EBR) systems. However, after long period practice Taobao, we find that...

In many classical e-commerce platforms, personalized recommendation has been proven to be of great business value, which can improve user satisfaction and increase the revenue platforms. this paper, we present a new problem, Trigger-Induced Recommendation (TIR), where users' instant interest explicitly induced with trigger item follow-up related target items are recommended accordingly. TIR become ubiquitous popular in figure out that although existing models effective traditional scenarios...

Click-Through Rate (CTR) prediction plays an important role in many industrial applications, such as online advertising and recommender systems. How to capture users' dynamic evolving interests from their behavior sequences remains a continuous research topic the CTR prediction. However, most existing studies overlook intrinsic structure of sequences: are composed sessions, where sessions user behaviors separated by occurring time. We observe that highly homogeneous each session,...

Conversion Rate (CVR) prediction in modern industrial e-commerce platforms is becoming increasingly important, which directly contributes to the final revenue. In order address well-known sample selection bias (SSB) and data sparsity (DS) issues encountered during CVR modeling, abundant labeled macro behaviors (i.e., user's interactions with items) are used. Nonetheless, we observe that several purchase-related micro specific components on item detail page) can supplement fine-grained cues...

A good personalized product search (PPS) system should not only focus on retrieving relevant products, but also consider user preference. Recent work PPS mainly adopts the representation learning paradigm, e.g., representations for each entity (including user, and query) from historical behaviors (aka. user-product-query interactions). However, we argue that existing methods do sufficiently exploit crucial collaborative signal, which is latent in interactions to reveal affinity between...

User preference modeling is a vital yet challenging problem in personalized product search. In recent years, latent space based methods have achieved state-of-the-art performance by jointly learning semantic representations of products, users, and text tokens. However, existing are limited their ability to model user preferences. They typically represent users the products they visited short span time using attentive models lack exploit relational information such as user-product...

Waterfall Recommender System (RS), a popular form of RS in mobile applications, is stream recommended items consisting successive pages that can be browsed by scrolling. In waterfall RS, when user finishes browsing page, the edge (e.g., phones) would send request to cloud server get new page recommendations, known as paging mechanism. RSs typically put large number into one reduce excessive resource consumption from numerous requests, which, however, diminish RSs' ability timely renew...

Capturing users' precise preferences is a fundamental problem in large-scale recommender system. Currently, item-based Collaborative Filtering (CF) methods are common matching approaches industry. However, they not effective to model dynamic and evolving of users. In this paper, we propose new sequential deep (SDM) capture by combining short-term sessions long-term behaviors. Compared with existing sequence-aware recommendation methods, tackle the following two inherent problems real-world...

E-commerce search engines comprise a retrieval phase and ranking phase, where the first one returns candidate product set given user queries. Recently, vision-language pre-training, combining textual information with visual clues, has been popular in application of tasks. In this paper, we propose novel V+L pre-training method to solve problem Taobao Search. We design task based on contrastive learning, outperforming common regression-based addition, adopt two negative sampling schemes,...

Recommender system (RS) devotes to predicting user preference a given item and has been widely deployed in most web-scale applications. Recently, knowledge graph (KG) attracts much attention RS due its abundant connective information. Existing methods either explore independent meta-paths for user-item pairs over KG, or employ neural network (GNN) on whole KG produce representations users items separately. Despite effectiveness, the former type of fails fully capture structural information...

Click-through rate (CTR) prediction is a critical task for many industrial systems, such as display advertising and recommender systems. Recently, modeling user behavior sequences attracts much attention shows great improvements in the CTR field. Existing works mainly exploit mechanism based on embedding product when considering relations between behaviors target item. However, this methodology lacks of concrete semantics overlooks underlying reasons driving to click In paper, we propose new...

In e-commerce search engines, query rewriting (QR) is a crucial technique that improves shopping experience by reducing the vocabulary gap between user queries and product catalog. Recent works have mainly adopted generative paradigm. However, they hardly ensure high-quality generated rewrites do not consider personalization, which leads to degraded relevance. this work, we present Contrastive Learning Enhanced Query Rewriting (CLE-QR), solution used in Taobao search. It uses novel...

Product search is an important service on Taobao, the largest e-commerce platform in China. Through this service, users can easily find products relevant to their specific needs. Coping with billion-size query loads, Taobao product has traditionally relied classical term-based retrieval models due powerful and interpretable indexes. In essence, efficient hinges proper storage of inverted index. Recent successes involve reducing size (pruning) index but construction deployment lossless static...

Recommender system, as an essential part of modern e-commerce, consists two fundamental modules, namely Click-Through Rate (CTR) and Conversion (CVR) prediction. While CVR has a direct impact on the purchasing volume, its prediction is well-known challenging due to Sample Selection Bias (SSB) Data Sparsity (DS) issues. Although existing methods, typically built user sequential behavior path ``impression$\to$click$\to$purchase'', effective for dealing with SSB issue, they still struggle...

Recommender Systems (RS), as an efficient tool to discover users' interested items from a very large corpus, has attracted more and attention academia industry. As the initial stage of RS, large-scale matching is fundamental yet challenging. A typical recipe learn user item representations with two-tower architecture then calculate similarity score between both representation vectors, which however still struggles in how properly deal negative samples. In this paper, we find that common...

E-commerce search engines comprise a retrieval phase and ranking phase, where the first one returns candidate product set given user queries. Recently, vision-language pre-training, combining textual information with visual clues, has been popular in application of tasks. In this paper, we propose novel V+L pre-training method to solve problem Taobao Search. We design task based on contrastive learning, outperforming common regression-based addition, adopt two negative sampling schemes,...

A good personalized product search (PPS) system should not only focus on retrieving relevant products, but also consider user preference. Recent work PPS mainly adopts the representation learning paradigm, e.g., representations for each entity (including user, and query) from historical behaviors (aka. user-product-query interactions). However, we argue that existing methods do sufficiently exploit crucial collaborative signal, which is latent in interactions to reveal affinity between...

Conversion Rate (\emph{CVR}) prediction in modern industrial e-commerce platforms is becoming increasingly important, which directly contributes to the final revenue. In order address well-known sample selection bias (\emph{SSB}) and data sparsity (\emph{DS}) issues encountered during CVR modeling, abundant labeled macro behaviors ($i.e.$, user's interactions with items) are used. Nonetheless, we observe that several purchase-related micro specific components on item detail page) can...