Research On The Algorithm Of IPTV Live Channel Recommendation Based On Deep Learning

With the fast development of Internet technology and the rapid increase of network bandwidth,Internet Protocol TV(IPTV)can provide more live TV channels and other diversified content services than traditional TV,so it has become more popular among home users.However,IPTV users also have to face the problem of information overload,and they have to spend more time searching for live TV channels that satisfy their interests.Electronic Program Guide(EPG)is still a common auxiliary tool provided by IPTV service providers to help users select channels,but EPG only provides a list containing all channel program description information in a hierarchical menu,and the burden of decision-making is still borne by users.With the success of personalized recommendation technology in Video on Demand and OTT business,in order to improve user experience,this paper studies the personalized recommendation system for IPTV live channels.The main contents of the work are summarized as follows:First,based on the sequential data of users watching IPTV live channels,this paper proposes a fusion recommendation method for IPTV live channels based on self-attention mechanism(F-TASICR).The method first constructs a temporal self-attention network to adaptively capture the channel switching patterns in the user’s historical viewing records,and then uses four basic recommenders based on channel viewing feature statistical strategies to extract implicit features,such as viewing duration,switching frequency,etc.Then,the two modules are integrated through the attention mechanism to jointly generate channel recommendation lists for users.Experimental results on a real large-scale IPTV dataset demonstrate the effectiveness of the proposed method in IPTV live channel recommendation scenarios.Second,taking the complex characteristics of home users behind IPTV devices in real scenarios into consideration,a recommendation method for IPTV live channels based on user and time grouping is proposed.This method dynamically groups users or time through the user’s viewing feature vector and the time distribution of the overall data in the dataset,thereby training a more fine-grained model and further improving the performance of short recommendation lists.Third,aiming at user historical behavior data and user interest characteristics,a general IPTV live channel recommendation method(LSICR)integrating long-term and short-term interests is proposed.The method first extracts the user’s long-term static interests through a temporal self-attention network,then uses a convolutional neural network with multiple filters of different sizes to extract the user’s short-term dynamic interests,and then simultaneously integrates the user’s long-term and short-term interests to generate personalized recommendation lists.Finally,experimental results on IPTV dataset and several public datasets demonstrate the effectiveness of the proposed method.