Figure - available from: International Journal of Machine Learning and Cybernetics
This content is subject to copyright. Terms and conditions apply.
![General multi-task learning model framework. The Shared-Bottom network is usually at the bottom, denoted as f, and multiple tasks share this layer. Up, the K subtasks correspond to a tower network, denoted as hK\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_{K}$$\end{document}, and the output of each subtask is yK=hK(f(x))\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{K}=h_{K}(f(x))$$\end{document}](publication/356191078/figure/fig3/AS:1142412814618625@1649383848395/General-multi-task-learning-model-framework-The-Shared-Bottom-network-is-usually-at-the.png)
General multi-task learning model framework. The Shared-Bottom network is usually at the bottom, denoted as f, and multiple tasks share this layer. Up, the K subtasks correspond to a tower network, denoted as hK\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_{K}$$\end{document}, and the output of each subtask is yK=hK(f(x))\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{K}=h_{K}(f(x))$$\end{document}
Source publication
In the recommender system, the user’s historical behavior data is one of the most important sources of the system’s input data. According to the user’s feedback mechanism, behavior data can be divided into explicit feedback data and implicit feedback data. However, most recommendation algorithms focus separately on explicit feedback or implicit fee...
Similar publications
The recommendation of points of interest (POIs) is essential in location-based social networks. It makes it easier for users and locations to share information. Recently, researchers tend to recommend POIs by treating them as large-scale retrieval systems that require a large amount of training data representing query-item relevance. However, gathe...
Citations
... Instead of randomly initializing the user's representations, this method also analyzed the user's review records to enrich the representations. Long et al. [29] constructed a multi-task model to analyze the user's historical behavior data. This method divided the behavior data into explicit feedback and implicit feedback, instead of separately analyzing these two types of data, it combined these data as the multi-task model to make to final prediction. ...
With the explosive growth of online information, users may also face information overload. To handle this problem, recommender systems have become an effective strategy, which can analyze the characters of users and items to provide valuable information. One of the important types of information is the item’s side information. For example, in Amazon dataset, side information mainly includes visual side information (e.g., image and video), textual side information (e.g., title and description), and auxiliary side information (e.g., brand and category). To analyze various types of side information, some research designed multiple modalities for different types of side information, which can improve the performance of the recommender system. To analyze the deeper relationships between users and items, recent works also use a graph structure to represent the interactions. Existing works on multi-modal recommender systems using graph neural networks largely depend on the interaction records, while little effort focuses on the relationships between interactions and various types of side information. In this paper, we propose a novel multi-task learning model. First, we construct the interaction records to graphs for each modality to gather the representations, and then we analyze the representations of each modality and the specific side information based on the similarities. We design a Multi-task Multi-modal Graph Neural Network (MTMM-GNN) framework built upon message-passing with the attention mechanism of graph neural networks, which can generate the representations of users and items from interaction records, and then analyze the relationships between the representations from GNNs and item’s side information. We conduct experiments on two public datasets, Amazon and MovieLens, and the results of our model outperform the state-of-the-art methods.
... Collaborative filtering is the process of using the collaboration of users, to make a recommendation [21]. The way collaborative filtering works is as described, − ...
Nowadays, e-commerce is becoming an essential part of business for many reasons, including the simplicity, availability, richness and diversity of products and services, flexibility of payment methods and the convenience of shopping remotely without losing time. These benefits have greatly optimized the lives of users, especially with the technological development of mobile devices and the availability of the Internet anytime and anywhere. Because of their direct impact on the revenue of e-commerce companies, recommender systems are considered a must in this field. Recommender systems detect items that match the customer's needs based on the customer's previous actions and make them appear in an interesting way. Such a customized experience helps to increase customer engagement and purchase rates as the suggested items are tailored to the customer's interests. Therefore, perfecting recommendation systems that allow for more personalized and accurate item recommendations is a major challenge in the e-marketing world. In our study, we succeeded in developing an algorithm to suggest personal recommendations to customers using association rules via the Frequent-Pattern-Growth algorithm. Our technique generated good results with a high average probability of purchasing the next product suggested by the recommendation system.
... By jointly optimizing these tasks, the model can improve the buffer utilization and ensure smoother playback without interruptions. The third is in Quality Evaluation and Reinforcement Learning [20)]. Multi-task learning can incorporate a quality evaluation task, where the model learns to assess the quality of the streaming experience based on user feedback or objective metrics. ...
This paper presents a framework with a taxonomy for multi-task learning in the context of Dynamic Adaptive Streaming over HTTP (DASH). DASH is a widely used technology for video streaming, and multi-task learning has emerged as a promising approach to enhance the performance and user experience of DASH systems by jointly optimizing multiple related tasks. The framework provides a structured approach to design, train, and evaluate multi-task learning models in DASH, while the taxonomy categorizes the key components and approaches within the framework. The taxonomy includes task types, multitask learning approaches, input features, and training strategies. Task types encompass video quality adaptation, buffer management, bandwidth estimation, content pre-fetching, and resource allocation, representing the specific tasks involved in DASH. Multi-task learning approaches encompass methodologies such as shared representation learning, task-specific layers, multi-head architectures, knowledge distillation, and reinforcement learning, offering flexibility in model design and optimization. Input features cover video characteristics, network conditions, device capabilities, and user preferences, providing the necessary information for informed decision-making across tasks. Training strategies include joint training, alternate training, hierarchical training, task weighting, and task balancing, determining how the multi-task learning model is trained and optimized. By following the presented framework and taxonomy, researchers and practitioners can systematically approach the design, training, and evaluation of multi-task learning models in DASH. The framework enables the development of efficient and adaptive video streaming systems by leveraging the interdependencies among tasks. The taxonomy helps organize the components and approaches within the framework, aiding in a better understanding of the various aspects of multi-task learning in the DASH context. Overall, this framework and taxonomy provide a valuable resource for advancing the field of multitask learning in the dynamic and complex domain of video streaming over HTTP.
... The traditional CF technology has achieved great success in recommendation systems, however, there are still some problems that limit its application in the large-capacity, complex and growing data environment. One problem is data sparsity, also called cold start [14], because most recommendation system involves large-volume, complex, and growing data [15,16]. Therefore, how to accurately and efficiently extract useful information and predict user Donghong Tian, Qianqian Wu and Min Li have contributed equally to this work. ...
The recommendation system is a good choice to provide users with personalized services. Facing with a large amount of data, clustering technology can group similar users into one category, which can make more rapid and accurate recommendations for target users. Due to user’s interests changing over time, how to make reasonable recommendations is an important issue. In this article, an improved k-means clustering algorithm based on cuckoo search is proposed and an improved time correlation coefficient is added to the algorithm to improve the accuracy of recommendation system. The proposed clustering algorithm integrates intra-cluster compactness and inter-cluster separation, which can improve the similarity of users in the same cluster. The proposed time correlation coefficient creatively considers both the inherent connection between user’s preference and time, and the impact of the periodicity and continuity of time on user rating patterns.
... The main objective of the TR layer is to learn an unknown parameter , a worker's ability to complete any specific task. Pertaining to our model requirements, we adopt a NCF approach to learn user ability because of its ability to learn implicit preferences and also because it has been extensively studied for and applied to the reality recommender systems [25][26][27]. ...
... Decision-Focused Learning-NCF: We denote this model as DFL-NCF which is a decision focused learning method taken from the study [27] with NCF in the TR layer and takes only workers and tasks as input features. The input features are based on the NCF architecture. ...
Continuous timely repair and replacement of infrastructures, equipment and utilities play an important role in maintaining the smooth-running of a city or local community. Thereby, to help individuals and businesses go about their daily activities with ease, it is vital to develop a proper method for automatically identifying and assigning capable workers for tasks. This paper defines the community management service task allocation problem as CMS-TAP and hence an end-to-end “recommendation + allocation” network, i.e. a task recommender and allocation optimization network (denoted as TROpt-NET), is then developed for handling such problem. TROpt-NET consists of two layers, namely one for predicting worker ability and the other for allocating tasks which are TR Layer and TA Layer, corresponding to “recommendation” and “allocation” of tasks. Different from operations research approaches where workers are assigned to jobs based on their pre-labelled skills and fixed locations, we propose a task recommender and allocation optimization network. The TR layer is a task recommender system designed to learn implicit worker abilities for different tasks using Neural Collaborative Filtering (NCF) by mining a historical dataset of worker task completion. Whereas in the TA layer a differential optimization approach for allocation is used because of its differentiable property and ability to allow for backpropagation to the prediction layer. In this study, we first formulate the CMS-TAP problem as a recommendation +optimization problem and then propose and end-to-end network architecture that tackles the problem in a real-world setting. TROpt-NET curbs uncertainty and assumptions in optimization by learning to more accurately approximate worker ability across different tasks. Additionally, the network can learn implicit worker abilities enabling optimal utilization of workers across a wide range of tasks, which is often ignored in task allocation problems. We find that normalizing worker ability across all tasks improves the implicit learning capability of the network and that good approximations don’t always lead to optimal allocation but learning allocations by backpropagating through recommendations improves the allocation objective. Offline experiments on a real-world large-scale dataset demonstrate the effectiveness of our proposed TROpt-NET.
... After network training and curve fitting, the mapping of fuzzy control rules is completed. The fuzzy rules are stored in the neural network and packaged into a neural network module for the calculation of network demand [20]. ...
In order to solve the modeling problem of travel as a service (MaaS) collaborative dispatching system of railway passenger transport hub based on neural network algorithm, meet people’s needs, make up for the lack of high traffic travel pressure, and improve people’s living standards, through 30 random questionnaires, it is found that 28 people think that travel convenience improves their quality of life, and 2 people think that owning a car has little effect on travel convenience. Travel has always been people’s basic living needs. In recent years, with the improvement of people’s living standards, there are more and more urban vehicles. At the same time, the increase of vehicles also directly leads to the increasing pressure of urban traffic travel, and the problems of vehicle emission pollution and traffic congestion are becoming more and more obvious. With the national low-carbon environmental protection policies, green transportation has become the theme of the times. With the continuous development of shared transportation mode and intelligent information technology, the new transportation concept of “Mobility as a Service” based on this technical mode will highly integrate the existing transportation modes and travel services. With the support of MaaS system, people’s travel modes will change greatly. Starting with the MaaS cooperative dispatching system of railway passenger transport hub, a multiparameter fuzzy neural network control system dispatching algorithm is proposed to better help the modeling of MaaS cooperative dispatching system of railway passenger transport hub.
... where denotes the outer-loop learning rate. When is set to 1, DN will degrade to Alternate Training in MTL [23], which could corrupt the performance as shown in the parameters analysis (section 5.4). ...
Large-scale e-commercial platforms in the real-world usually contain various recommendation scenarios (domains) to meet demands of diverse customer groups. Multi-Domain Recommendation (MDR), which aims to jointly improve recommendations on all domains, has attracted increasing attention from practitioners and researchers. Existing MDR methods often employ a shared structure to leverage reusable features for all domains and several specific parts to capture domain-specific information. However, data from different domains may conflict with each other and cause shared parameters to stay at a compromised position on the optimization landscape. This could deteriorate the overall performance. Despite the specific parameters are separately learned for each domain, they can easily overfit on data sparsity domains. Furthermore, data distribution differs across domains, making it challenging to develop a general model that can be applied to all circumstances. To address these problems, we propose a novel model agnostic learning method, namely MAMDR, for the multi-domain recommendation. Specifically, we first propose a Domain Negotiation (DN) strategy to alleviate the conflict between domains and learn better shared parameters. Then, we develop a Domain Regularization (DR) scheme to improve the generalization ability of specific parameters by learning from other domains. Finally, we integrate these components into a unified framework and present MAMDR which can be applied to any model structure to perform multi-domain recommendation. Extensive experiments on various real-world datasets and online applications demonstrate both the effectiveness and generalizability of MAMDR.
Social media has found its path into the daily lives of people. There are several ways that users communicate in which liking and sharing images stands out. Each image shared by a user can be analyzed from aesthetic and personality traits views. In recent studies, it has been proved that personality traits impact personalized image aesthetics assessment. In this article, the same pattern was studied from a different perspective. So, we evaluated the impact of image aesthetics on personality traits to check if there is any relation between them in this form. Hence, in a two-stage architecture, we have leveraged image aesthetics to predict the personality traits of users. The first stage includes a multi-task deep learning paradigm that consists of an encoder/decoder in which the core of the network is a Swin Transformer. The second stage combines image aesthetics and personality traits with an attention mechanism for personality trait prediction. The results showed that the proposed method had achieved an average Spearman Rank Order Correlation Coefficient (SROCC) of 0.776 in image aesthetic on the Flickr-AES database and an average SROCC of 0.6730 on the PsychoFlickr database, which outperformed related SOTA (State of the Art) studies. The average accuracy performance of the first stage was boosted by 7.02 per cent in the second stage, considering the influence of image aesthetics on personality trait prediction.
