scholarly journals Deep Match to Rank Model for Personalized Click-Through Rate Prediction

2020 ◽  
Vol 34 (01) ◽  
pp. 156-163 ◽  
Author(s):  
Zequn Lyu ◽  
Yu Dong ◽  
Chengfu Huo ◽  
Weijun Ren
Keyword(s):  
Recommender System ◽  
User Behavior ◽  
Target Item ◽  
Inner Product ◽  
User Interest ◽  
Ranking Task ◽  
Click Through Rate ◽  
Novel Model ◽  
State Of Art ◽  
Model Personalization

Click-through rate (CTR) prediction is a core task in the field of recommender system and many other applications. For CTR prediction model, personalization is the key to improve the performance and enhance the user experience. Recently, several models are proposed to extract user interest from user behavior data which reflects user's personalized preference implicitly. However, existing works in the field of CTR prediction mainly focus on user representation and pay less attention on representing the relevance between user and item, which directly measures the intensity of user's preference on target item. Motivated by this, we propose a novel model named Deep Match to Rank (DMR) which combines the thought of collaborative filtering in matching methods for the ranking task in CTR prediction. In DMR, we design User-to-Item Network and Item-to-Item Network to represent the relevance in two forms. In User-to-Item Network, we represent the relevance between user and item by inner product of the corresponding representation in the embedding space. Meanwhile, an auxiliary match network is presented to supervise the training and push larger inner product to represent higher relevance. In Item-to-Item Network, we first calculate the item-to-item similarities between user interacted items and target item by attention mechanism, and then sum up the similarities to obtain another form of user-to-item relevance. We conduct extensive experiments on both public and industrial datasets to validate the effectiveness of our model, which outperforms the state-of-art models significantly.

scholarly journals Deep Interest Evolution Network for Click-Through Rate Prediction

2019 ◽  
Vol 33 ◽  
pp. 5941-5948 ◽  
Author(s):  
Guorui Zhou ◽  
Na Mou ◽  
Ying Fan ◽  
Qi Pi ◽  
Weijie Bian ◽  
...  
Keyword(s):  
User Behavior ◽  
Target Item ◽  
State Of The Art ◽  
User Interest ◽  
User Interests ◽  
Sequential Structure ◽  
Changing Trend ◽  
Click Through Rate ◽  
Novel Model ◽  
Deep Interest

Click-through rate (CTR) prediction, whose goal is to estimate the probability of a user clicking on the item, has become one of the core tasks in the advertising system. For CTR prediction model, it is necessary to capture the latent user interest behind the user behavior data. Besides, considering the changing of the external environment and the internal cognition, user interest evolves over time dynamically. There are several CTR prediction methods for interest modeling, while most of them regard the representation of behavior as the interest directly, and lack specially modeling for latent interest behind the concrete behavior. Moreover, little work considers the changing trend of the interest. In this paper, we propose a novel model, named Deep Interest Evolution Network (DIEN), for CTR prediction. Specifically, we design interest extractor layer to capture temporal interests from history behavior sequence. At this layer, we introduce an auxiliary loss to supervise interest extracting at each step. As user interests are diverse, especially in the e-commerce system, we propose interest evolving layer to capture interest evolving process that is relative to the target item. At interest evolving layer, attention mechanism is embedded into the sequential structure novelly, and the effects of relative interests are strengthened during interest evolution. In the experiments on both public and industrial datasets, DIEN significantly outperforms the state-of-the-art solutions. Notably, DIEN has been deployed in the display advertisement system of Taobao, and obtained 20.7% improvement on CTR.

A Dynamic Recommender System with Fused Time and Location Factors

2019 ◽  
Vol 23 (1) ◽  
pp. 25-33
Author(s):  
Xinhua Wang ◽  
Peng Yin ◽  
Yukai Gao ◽  
Lei Guo ◽  
◽  
...  
Keyword(s):  
Recommender System ◽  
User Behavior ◽  
Geographical Location ◽  
User Preferences ◽  
Dynamic Features ◽  
Fusion Model ◽  
User Interest ◽  
Location Factors ◽  
User Behavior Analysis ◽  
Real World Datasets

A recommender system is an important tool to help users obtain content and overcome information overload. It can predict users’ interests and offer recommendations by analyzing their history behaviors. However, traditional recommender systems focus primarily on static user behavior analysis. Recently, with the promotion of the Netflix recommendation prize and the open dataset with location and time information, many researchers have focused on the dynamic characteristics of the recommender system (including the changes in the dynamic model of user interest), and begun to offer recommendations based on these dynamic features. Intuitively, these dynamic user features provide us with an effective method to learn user interests deeply. Based on the observations above, we present a dynamic fusion model by integrating geographical location, user preferences, and the time factor based on the Gibbs sampling process to provide better recommendations. To evaluate the performance of our proposed method, we conducted experiments on real-world datasets. The experimental results indicate that our proposed dynamic recommender system with fused time and location factors not only performs well in traditional scenarios, but also in sparsity situations where users appear at the first time.

scholarly journals AANMF: Attribute-Aware Attentional Neural Matrix Factorization

2019 ◽  
Vol 48 (4) ◽  
pp. 682-693
Author(s):  
Bo Zheng ◽  
Jinsong Hu
Keyword(s):  
Matrix Factorization ◽  
Recommendation System ◽  
Auxiliary Information ◽  
Inner Product ◽  
Data Sets ◽  
It Projects ◽  
Real World Data ◽  
Latent Space ◽  
Almost All ◽  
Novel Model

Matrix Factorization (MF) is one of the most intuitive and effective methods in the Recommendation System domain. It projects sparse (user, item) interactions into dense feature products which endues strong generality to the MF model. To leverage this interaction, recent works use auxiliary information of users and items. Despite effectiveness, irrationality still exists among these methods, since almost all of them simply add the feature of auxiliary information in dense latent space to the feature of the user or item. In this work, we propose a novel model named AANMF, short for Attribute-aware Attentional Neural Matrix Factorization. AANMF combines two main parts, namely, neural-network-based factorization architecture for modeling inner product and attention-mechanism-based attribute processing cell for attribute handling. Extensive experiments on two real-world data sets demonstrate the robust and stronger performance of our model. Notably, we show that our model can deal with the attributes of user or item more reasonably. Our implementation of AANMF is publicly available at https://github.com/Holy-Shine/AANMF.

scholarly journals Deep Time-Stream Framework for Click-through Rate Prediction by Tracking Interest Evolution

2020 ◽  
Vol 34 (04) ◽  
pp. 5726-5733
Author(s):  
Shu-Ting Shi ◽  
Wenhao Zheng ◽  
Jun Tang ◽  
Qing-Guo Chen ◽  
Yao Hu ◽  
...  
Keyword(s):  
Continuous Time ◽  
Industrial Applications ◽  
Superior Performance ◽  
User Interest ◽  
Additional Time ◽  
Deep Time ◽  
Public Dataset ◽  
Click Through Rate ◽  
Time Information ◽  
Video Recommendation

Click-through rate (CTR) prediction is an essential task in industrial applications such as video recommendation. Recently, deep learning models have been proposed to learn the representation of users' overall interests, while ignoring the fact that interests may dynamically change over time. We argue that it is necessary to consider the continuous-time information in CTR models to track user interest trend from rich historical behaviors. In this paper, we propose a novel Deep Time-Stream framework (DTS) which introduces the time information by an ordinary differential equations (ODE). DTS continuously models the evolution of interests using a neural network, and thus is able to tackle the challenge of dynamically representing users' interests based on their historical behaviors. In addition, our framework can be seamlessly applied to any existing deep CTR models by leveraging the additional Time-Stream Module, while no changes are made to the original CTR models. Experiments on public dataset as well as real industry dataset with billions of samples demonstrate the effectiveness of proposed approaches, which achieve superior performance compared with existing methods.

A Method of Interest Degree Mining Based on Behavior Data Analysis

2019 ◽  
Vol 34 (09) ◽  
pp. 2059030
Author(s):  
Zhen Li ◽  
Shuo Xu ◽  
Tianyu Wang
Keyword(s):  
Social Media ◽  
Big Data ◽  
Data Fusion ◽  
Language Processing ◽  
User Behavior ◽  
Comprehensive Evaluation ◽  
Mapping Method ◽  
User Preference ◽  
User Interest ◽  
Improvement Measures

Based on big data, this paper starts from the behavior data of users on social media, and studies and explores the core issues of user modeling under personalized services. Focusing on the goal of user interest modeling, this paper proposes corresponding improvement measures for the existing interest model, which has great difference in interest description among different users and it is difficult to find the user interest change in time. For the above problems, this paper takes user-generated content and user behavior information as the analysis object, and uses natural language processing, knowledge warehouse, data fusion and other methods and techniques to numerically analyze user interest mining based on text mining and multi-source data fusion. We propose a user interest label space mapping method to avoid data sparse problem caused by too many dimensions in interest analysis. At the same time, we propose a method to extract and blend the long-term and short-term interests, and realize the comprehensive evaluation of interests. In the analysis of the big data phase, the user preference social property application preference value law, it is expected to achieve user Internet social media application preference data mining from the perspective of big data.

scholarly journals TransGate: Knowledge Graph Embedding with Shared Gate Structure

2019 ◽  
Vol 33 ◽  
pp. 3100-3107 ◽  
Author(s):  
Jun Yuan ◽  
Neng Gao ◽  
Ji Xiang
Keyword(s):  
Time Complexity ◽  
Link Prediction ◽  
Graph Embedding ◽  
Feature Engineering ◽  
Specific Information ◽  
Gate Structure ◽  
Essential Problem ◽  
Novel Model ◽  
State Of Art ◽  
Parameter Sharing

Embedding knowledge graphs (KGs) into continuous vector space is an essential problem in knowledge extraction. Current models continue to improve embedding by focusing on discriminating relation-specific information from entities with increasingly complex feature engineering. We noted that they ignored the inherent relevance between relations and tried to learn unique discriminate parameter set for each relation. Thus, these models potentially suffer from high time complexity and large parameters, preventing them from efficiently applying on real-world KGs. In this paper, we follow the thought of parameter sharing to simultaneously learn more expressive features, reduce parameters and avoid complex feature engineering. Based on gate structure from LSTM, we propose a novel model TransGate and develop shared discriminate mechanism, resulting in almost same space complexity as indiscriminate models. Furthermore, to develop a more effective and scalable model, we reconstruct the gate with weight vectors making our method has comparative time complexity against indiscriminate model. We conduct extensive experiments on link prediction and triplets classification. Experiments show that TransGate not only outperforms state-of-art baselines, but also reduces parameters greatly. For example, TransGate outperforms ConvE and RGCN with 6x and 17x fewer parameters, respectively. These results indicate that parameter sharing is a superior way to further optimize embedding and TransGate finds a better trade-off between complexity and expressivity.

scholarly journals The Impact of User Behavior on Information Diffusion in D2D Communications: A Discrete Dynamical Model

2018 ◽  
Vol 2018 ◽  
pp. 1-9 ◽  
Author(s):  
Chenquan Gan ◽  
Xiaoke Li ◽  
Lisha Wang ◽  
Zufan Zhang
Keyword(s):  
Information Diffusion ◽  
User Behavior ◽  
Transition Process ◽  
Dynamical Model ◽  
Stability Factor ◽  
User Interest ◽  
User Influence ◽  
Proposed Model ◽  
Interaction Factor ◽  
The Impact

This paper aims to explore the impact of user behavior on information diffusion in D2D (Device-to-Device) communications. A discrete dynamical model, which combines network metrics and user behaviors, including social relationship, user influence, and interest, is proposed and analyzed. Specifically, combined with social tie and user interest, the success rate of data dissemination between D2D users is described, and the interaction factor, user influence, and stability factor are also defined. Furthermore, the state transition process of user is depicted by a discrete-time Markov chain, and global stability analysis of the proposed model is also performed. Finally, some experiments are examined to illustrate the main results and effectiveness of the proposed model.

Sign in / Sign up

Export Citation Format

Share Document