Research On Ancient Poem Recommendation Algorithm Integrating Emotional Knowledge

Ancient poetry,as a learning focus of Chinese education in primary and middle schools,plays a vital role in exam-oriented education and quality education system.As an intelligence knowledge service,ancient poetry recommendation can actively recommend students similar poetry when they expand ancient poetry learning.Since emotion containing in poetry is one of the most important content in poetry learning,this paper will take emotion as the most important dimension feature in ancient poetry recommendation.And the basis of this fact it will study on ancient poetry recommendation based on emotion classification.The contributions are summarized as follows:This paper proposes an emotion classification model based on imagery dictionary(IIK-ECM).Since ancient poetry texts are always obscure and complicated and they are quite different from contemporary Chinese,conventional emotion classification model may perform badly due to difficulty of comprehension.In order to alleviate this problem,this paper firstly study traditional imagery connotation in Chinese ancient poetry and excavate emotion representation containing in imagery.And then,an imagery dictionary is constructed.IIK-ECM uses the image dictionary to incorporate the emotional content corresponding to the image words in the ancient poetry text.The model classifies ancient poetry texts that incorporate emotional knowledge,improving the effect of ancient poetry emotion classification.In specific experiments,this paper verifies the accuracy of imagery dictionary construction using Three Hundred Tang Poems data set in an artificial method,and then contrast experiments are done on FSPC,an emotion intensity data set.The results show that IIK-ECM in this paper is better than baseline model both on accuracy and F1 score of emotion classification evaluation metrics.In order to enlarge recommendation dimensions,improve the recommendation accuracy on emotion dimension and make the results more reliable,this paper proposes a recommendation method fused with emotion knowledge.Firstly,this method extracts emotion knowledge with the above IIK-ECM.And then,ancient poetry will be represented as vectors using a TransE model based on Bernoulli Negative Sampling Algorithm(BNS-TransE).Since the generated embedding vectors contain emotion knowledge,ancient poetry will be better represented in emotion dimension.Finally,this method gets the similarity by computing Euclidean distance between vectors and generates a recommendation list.In detail,contrast experiments of BNS-TransE have been done on poetry knowledge database,whose results shows that this model performs better than baseline model on knowledge representation evaluation metric such as MR and HITS1.Based on the result in ancient poetry recommendation experiment,in the recommendation list fused with emotion knowledge,the number of ancient poetry has significantly increased,whose emotion character is the same as the goal poetry's.It is proved that recommendation results fused with emotion knowledge are more reliable and more accurate in emotion dimension.