Providing effective recommendations in discussion groups using a new hybrid recommender system based on implicit ratings and semantic similarity
Introduction
Various communicative environments in different domains have been provided for users via the Internet platform using different web technologies. Asynchronous discussion groups are important examples of these communicative environments which allow users to find proper answers for their raised questions (Kardan and Ebrahimi, 2013). Knowledge extraction from discussion groups and communities is an important research issue that is considered in a variety of recent studies (Kaššák et al., 2016, Christensen and Schiaffino, 2011, Ortega et al., 2016, Wang et al., 2016, Jhamb and Fang, 2017, Lee and Brusilovsky, 2017, Xu, 2018, Li et al., 2018). Unstructured nature of the posts of the members and large volume of information are the main problems of the knowledge extraction process from discussion groups. Recommender systems can be exploited to extract useful knowledge from discussion groups.
This study aims to deliver the appropriate contents that are posted by members of discussion groups to the inquirer users. For this purpose, the related contents are identified at the first step and then, according to the member's interests, the appropriate recommendation will be provided. Most of recommender systems use the similarity of users to make recommendations in discussion groups. Considering similarities of content, or user and content, for this purpose has not received much attention in these systems. The structure of discussion groups is similar to a tree structure and their contents can be as main groups, subgroups, discussion topics, and posts (Fanaeetork and Yazdi, 2013). Each discussion in the subgroups is called a “thread” which has the date and subject by default. This structure allows users to search related topics more quickly. There are some drawbacks in this structure. For example, the users cannot search whole of the group when one topic is included at several threads and while the size of the group is extra-large and thus, they cannot find their related information. Since users cannot usually explain their interests using keyword-based queries. In these situations, common keyword-based searching of discussion groups cannot works properly. Recommender systems work based on users’ activities, behaviors, and preferences.
Sparsity and cold start problems have made collaborative filtering techniques inappropriate for developing recommendations in discussion groups. The semantic of each post should be discovered in order to increase the quality of recommendations in discussion groups. Therefore, in content-based filtering techniques, in order to extraction of related content, the semantic similarity should be applied instead of other methods which use keyword-based measures. Furthermore, the contents of the posts are ignored by the collaborative filtering recommender systems of discussion groups. A proper solution to overcome the mentioned challenges is to use hybrid techniques which consider information of both user and content. The hybrid recommender system is rarely presented in discussion groups’ domain.
In this study, we are going to resolve the mentioned challenges. For this purpose, a new hybrid recommendation technique for discussion groups is suggested. Since the users’ explicit information is usually unavailable, their implicit information should be used. Some ideas of collaborative filtering part of the proposed technique are based on activity analysis and user’s behavior in discussion groups, which retrieve the user's implicit information. Based on this implicit information, some functions are introduced to calculate implicit ratings of posts. Similar users are identified according to the extracted implicit ratings. The semantic similarity of posts is the basis of suggested plan of the content-based filtering part of the proposed method. For this purpose, the present tags in each subgroup should be identified separately and they should be also organized according to their semantic relationships. Then, during searching for the similar posts, these identified semantic relationships are used and related tags are discovered and similar posts are found based on them. The results of two mentioned parts are combined in hybrid filtering part and the final recommendations are made in the last step.
The main contributions of this paper can be summarized as follows:
- •
A new hybrid method is proposed to provide recommendation in discussion groups. This method includes 3 main parts, namely, collaborative filtering part, content-based filtering part, and hybrid filtering part.
- •
In collaborative filtering part of the proposed method, the users’ implicit ratings obtained based on presented ideas and users which are similar to the active user are identified.
- •
In content-based filtering part, similar posts to the user’s query are identified according to the semantic relevance between user’s question and existing posts.
- •
Finally, in the hybrid filtering part a relation is provided in which obtained posts are recommended to active user based on what extent similar users to the active user have contributed in them.
The remaining of this paper is organized as follows: In Section 2, related works are explained. Section 3 introduces the proposed method. In Section 4, the experimental results and their related evaluation and analytical discussions are described. Finally, the work is concluded in Section 5.
Section snippets
Related works
In this section, different types of recommender systems are explained first and some of provided studies are reviewed and then, some examples of discussion groups are given. Moreover, the employed techniques in suggested system are stated briefly.
Content-based filtering (Pera and Ng, 2013, Son and Kim, 2017, Boratto et al., 2017), collaborative filtering (Yang et al., 2014), and hybrid filtering (Yang et al., 2017; Xu, 2018) are the main filtering types of recommender systems (Adomavicius and
Proposed method
A few researches have been provided about using recommender systems in domain of discussion groups which consider both content information and user information in their recommendations. A hybrid recommender system is proposed in this section that consists of three main parts, namely, content-based filtering, collaborative filtering, and hybrid filtering. Furthermore, existing information in database of discussion groups is used in pre-processing section to organize existing tags to be exploited
Experimental results, evaluation and analysis
In this section, the approach of analyzing and evaluating the proposed method is explained and the obtained results of experiments will be demonstrated and analyzed.
Conclusion
Considering the features of discussion groups, a hybrid recommender system that consists of three parts, namely, content-based filtering, collaborative filtering, and hybrid filtering was proposed in this paper to enhance performance of Kardan and Ebrahimi (2013). This study was conducted based on discussion groups with tagging feature. Semantic relations between existing tags in each subgroup were obtained using WordNet dictionary and tags were organized in a hierarchical structure according
CRediT authorship contribution statement
Masoumeh Riyahi: Methodology, Software. Mohammad Karim Sohrabi: Writing - review & editing, Supervision, Conceptualization.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (103)
A new similarity measure for collaborative filtering to alleviate the new user cold-starting problem
Inf. Sci.
(2008)- et al.
Fuzzy-genetic approach to recommender systems based on a novel hybrid user model
Expert Syst. Appl.
(2008) - et al.
A hybrid content-based and item-based collaborative filtering approach to recommend TV programs enhanced with singular value decomposition
Inf. Sci.
(2010) - et al.
A framework for collaborative filtering recommender systems
Expert Syst. Appl.
(2011) - et al.
Collaborative filtering based on significances
Inf. Sci.
(2012) - et al.
Recommender systems survey
Knowl.-Based Syst.
(2013) - et al.
Semantics-aware content-based recommender systems: design and architecture guidelines
Neurocomputing
(2017) - et al.
A hybrid online-product recommendation system: combining implicit rating-based collaborative filtering and sequential pattern analysis
Electron. Commer. Res. Appl.
(2012) - et al.
Entertainment recommender systems for group of users
Expert Syst. Appl.
(2011) - et al.
Merging trust in collaborative filtering to alleviate data sparsity and cold star
Knowl.-Based Syst.
(2014)