【摘要】 为从影片简介中获取更丰富的文本特征并考虑短文本的语义依赖性,构建一种基于卷积双向长短记忆神经网络模型(CBiLSTM)的视频文本特征表示模型。根据用户评分行为习惯的相似性和偏向性,基于受偏置约束的概率矩阵分解算法设计一种混合视频推荐算法(CBiLSTM-PMF+),实现评分预测。在真实数据集上结合IMDB数据库视频简介进行实验仿真,实验结果表明了影片文本局部特征和上下文时序特征结合进行评分预测的合理性。更多还原

【Abstract】 To obtain richer text features from the movie introduction and consider the semantic dependency of short text,a video text feature representation model based on a convolutional bidirectional long and short memory neural network model(CBiLSTM)was constructed.Based on the similarity and bias of the user’s rating behavior habits,a hybrid video recommendation algorithm(CBiLSTM-PMF+)was designed based on the bias-constrained probability matrix decomposition algorithm to achieve rating prediction.Experimental simulation was performed on the real data set with the introduction of the IMDB database video.Experimental results verify the reasonableness of the score prediction by combining the local features of the movie text and the context temporal features.更多还原