Heterogeneous Information Network + Recommendation = = Summary _ Network analysis related

About the basic concept of heterogeneous information network, you can first look at: Heterogeneous Information network-Basic concepts and definitions learning notes

* * One, meta path (meta path)

**

-Because heterogeneous information networks (hereafter referred to as Hin) contain more types of nodes and edges, and are more complex than homogeneous networks, Yizhou Sun "1" first proposed the concept of Meta path (meta path), based on the meta path to analyze Hin, In simple terms, a meta path is a concatenation of the type of edge and the type of node that connects two nodes. A detailed view of heterogeneous information networks – Basic concepts and definitions of learning notes.

-The basic two jobs based on the meta path are:
(1) To measure the similarity between nodes of the same type based on symmetric element paths in Hin Pathsim "1" (Code python2.7:http://download.csdn.net/download/u013527419/9475257).
(2) Measure the similarity method between the same/different types of nodes based on arbitrary Hin in the Hetesim "2" (Code: https://download.csdn.net/download/u013527419/10353251).
Of course there are many others: Pathcount (such as this preference to the high visibility of the node is not more should be applied to the recommended task.) Intuition is also the choice of more popular items), PCRW and so on. Later, based on these two basic methods and have a lot of work, after all, the similarity is some like clustering, link prediction, recommendations and other work of the foundation. This aspect of interest can refer to the Beijing Posts and telecommunications Chuan Shi teacher Publications:http://dblp.uni-trier.de/pers/hd/s/shi:chuan.
(3) Other methods of similarity measurement based on meta path:
-Common NODE,PCRW,BPCRW
-Knowsim (based on given Meta path and reverse meta path)
-Avgsim (similarity of documents in HIN)
-Relsim (Measure the similarity of relations in HIN)

In

3.HIN, the recommended work based on Meta path is:
(1) Semrec "3" is recommended in weighted Hin by considering the similarity between user and item based on different meta Poth;
(2) Yu et al. "4 "The method of matrix decomposition is used to recommend the implicit characteristics of user and item based on different meta Poth;

Both methods have different results based on different meta path, so there is a problem with data fusion," 3 "with Hetesim. The similarity between user and item based on different meta path is obtained, and the similarity is combined with different weights, and this method does not consider implict factors, and the different similarity matrices used to do ensemble can be sparse. "4" is based on the matrix decomposition of the different meta Poth of user and item of the implicit vector representation, and then by assigning different weights to the inner product to fit the real score, to get the weight value. In the process of doing ensemble in the end, this method only takes into account the variables associated with the current predictive value, which is not related to other values.
This can be seen in Yizhou Sun's related article: Http://www.ccs.neu.edu/home/yzsun/Publications.htm
or Xiang ren:http:// Xren7.web.engr.illinois.edu/,x. Yu and so on.
* *
Two, meta Structure or meta Graph

**



1. Given the limited capacity of the meta path representation. For example, in DBLP, if we want to measure the relationship between the two authors of a published paper that contains the same subject word (i.e.,the APVPA and APTPA paths) at the same meeting, the meta path is not very good, such as A1 and A2, based on Meta The three kinds of similarity of path are the same. "5" brings up meta structure, which can represent some more complex relationships, and Meta path is a special case of meta structure.
2. The recommendation based on meta structure can look at "6".
Interested in this area can be seen: https://www.cse.ust.hk/~yqsong/and http://i.cs.hku.hk/~ckcheng/

Reference documents:
"1" Sun, y.z, Han, J.W, Yan, X.f, Yu, P.s., Wu, T.: Pathsim:meta path-based top-k Similarity
Search in heterogeneous information networks. In:vldb, pp. 992–1003 (2011)
"2" Shi, C., Kong, X., Huang, Y., Philip, S.y., Wu, B.: HETESIM:A general Framework for relevance
Measure in heterogeneous networks. IEEE Trans. Knowl. Data Eng 26 (10), 2479–2492 (2014)
"3" Shi, C., Zhang, Z., Luo, P., Yu, P.S, Yue, Y., Wu, B.: Semantic path based personalized
Recommendation on weighted heterogeneous information networks. IN:CIKM, pp. 453–462
(2015)
"4" Yu, X., Ren, X., Sun, Y., Gu, Q., Sturt, B., Khandelwal, U., Norick, B., Han, J.: Personalized
Entity recommendation:a heterogeneous Information network approach. IN:WSDM, pp. 283–
292 (2014)
"5" Huang, Z., Zheng, Y., Cheng, R., Sun, Y., Mamoulis, N., Li, X.: Meta structure:computing
Relevance in large heterogeneous information networks. IN:SIGKDD, pp. 1595–1604 (2016)
"6" Meta-graph Based recommendation Fusion over heterogeneous information Networks.

It feels like those people are in a big circle, and the last person to hang is Yizhou sun,philip S. Yu,jiawei Han.
Supplemental: Some messy summaries of the recommendation system can look at: http://write.blog.csdn.net/postlist/6424393/null