Research on Cross-Platform Image Recommendation Model Fusing Text Information

Table 2

Benchmarking checklist for methods of image recommendation.

Checklist issues references	Text information	Deep feature information	Image information
Sejal et al. [18]	(1) Recommendation based on the match of queried text and image description text.	—	—
	(2) A synonym term dictionary is used for extending text queries to improve recommendation.
Sejal et al. [19]	Recommendation based on the match of queried text and image annotation text.	—	—
Zhu et al. [20]	—	(1) Recommendation based on the match of deep features of queried images and candidate images.	—
		(2) Deep features of images are generated by using convolutional neural network.
Bo and Peng [21]	—	—	Recommendation based on the match of color histogram features or Gabor texture features of queried images and candidate images.
Widisinghe et al. [22]	(1) The user-generated tag information is applied to collaborative filtering for top-n recommendation.	—	c
Our method	(1) Single features of images are the image class labels generated by using convolutional neural network.	(1) Recommendation based on the match of word embeddings of queried text and fusion matrix.	—
	(2) Single features of text are keyword vectors generated by using keyword extraction method.	(2) The fusion matrix and queried text is transformed to the word embedding matrices by using Word2Vec.
	(3) Single features of images and text are fused to generate a fusion matrix with elements of “image class label — keyword” pairs.