Recommender systems still face a trade-off between exploring new items to maximize user satisfaction and exploiting those already interacted with match interests. This problem is widely recognized as the exploration/exploitation (EE) dilemma, multi-armed bandit (MAB) algorithm has proven be an effective solution. As scale of users in real-world application scenarios increases, their purchase interactions become sparser. Then three issues need investigated when building MAB-based recommender...

Image retrieval methods in the fashion field mainly take advantage of query images that reflect user needs, without considering additional keywords users can provide to specify attributes their interests. To achieve fine-grained retrieval, we propose an iterative similarity learning network (ISLN) for attribute-guided image which takes a and specified attribute as input, outputs other with same or similar values. The core is module, leverages aggressive ability deep neural (DNN) focus on...

Multi-behavior recommendation models excel in extracting abundant information from user-item interactions to enhance performance; however, they encounter challenges accuracy due noise disturbance and ambiguous weight allocation. In this paper, we propose cd-MBRec, a novel model designed amplify commonality among various behaviors, thereby minimizing interference while preserving behavior diversity highlight semantic variations feedback across distinct scenarios. Specifically, the begins by...

Clothing images vary in style and everyone has a different understanding of style. Even with the current popular deep learning methods, it is difficult to accurately classify labels. A representation model based on neural networks called StyleNet proposed this paper. We adopt multi-task framework build make full use various types label information represent clothing finer-grained manner. Due semantic abstraction image labels fashion field, using simple migration method cannot fully meet...

Fashion image retrieval is one of the important services e-commerce platforms, and it also basis various fashion-related AI applications. Studies have shown that in a multi-modal environment (images + attribute labels), embedding items into specific spaces can support more fine-grained similarity measures, which especially suitable for fashion tasks. In this paper, we propose an attention-based attribute-guided learning network (AttnFashion) retrieval. The core spatial attention module...

Sales forecasting is an important part of e-commerce and critical to smart business decisions. The traditional methods mainly focus on building a model, training the model through historical data, then using it forecast future sales. Such are feasible effective for products with rich data while they not performing as well newly listed little or no data. In this paper, idea collaborative filtering, similarity-based sales (S-SF) method proposed. implementation framework S-SF includes three...

To address the challenges of sparse data and high dynamics in Contexual Multi-Armed Bandits (CMAB) models for online recommendation, this study introduces a novel Knowledge Graph-driven Thompson Sampling (KG-TS) algorithm within CMAB framework. This innovatively constructs dynamic knowledge graph that links user characteristics to item attributes, converting sequential decision-making into structures explore relationships enhance contextual understanding. Notably, KG-TS incorporates...

There is a demand in the current fashion e-commerce field that users expect to find products are similar query image and match text description simultaneously. Traditional text-based retrieval image-based methods cannot deal with this issue. In paper, we propose fashion-oriented image-text representation learning model called FashionCorrNet. It an improvement correlational neural network (CorrNet) by deepening structure merging Kendall correlation coefficient Pearson objective function...

Matrix factorization (MF) models are effective and easy to expand widely used in industry, such as rating prediction item recommendation. The basic MF model is relatively simple. In practical applications, side information attributes or implicit feedback often combined improve accuracy by modifying the optimizing algorithm. this paper, we propose an attribute interaction-aware matrix (AIMF) method for recommendation tasks. We partition original into different sub-matrices according...

Learning the similarity between fashion items is essential for many fashion-related tasks. Most methods based on global or local image cannot meet fine-grained retrieval requirements related to attributes. We are first clearly distinguish concepts of attribute name and their values divide tasks that combine images text into: attribute-guided attribute-manipulated retrieval. propose a hierarchical attribute-aware embedding network (HAEN) takes attributes as input, learns multiple...

Extracting more information from feature interactions is essential to improve click-through rate (CTR) prediction accuracy. Although deep learning technology can help capture high-order interactions, the combination of features lacks interpretability. In this paper, we propose a multi-semantic interaction network (MeFiNet), which utilizes convolution operations map spaces their expressive ability and uses an improved Squeeze & Excitation method based on SENet learn importance these in...

New development of classification system and safety-protection strategy geologic survey data is summarized for searching optimized solutions disaster tolerance technology mass acquired in recent years by Chinese geological organizations. The new solution which building on the basis extensive research review empirical from national international best practices comparing pros cons backup technologies can provide some basic network supports social services data.