1. Introduction

2. Literature Review and Related Works

2.2. Approaches of News Recommendation System

The usual approaches used in the News Recommendation System (NRS) are three. These basic approaches of recommendation systems are Content-based Filtering, Collaborative filtering and Hybrid (which is a combination of the first two approaches). So, the details of these techniques were discussed.

2.2.1. Content-Based Filtering (CBF)

In our study, we have used different techniques as discussed above. From these techniques, one is the content-based filtering approach. Content-Based Filtering (CBF) is one of the techniques of a news recommendation system, which is based on the properties of the items and tries to recommend news items which are similar to those a given user has liked to read or rated in the past. This method finds the preferences of the current user about the articles of news using the rating history of the current user related to previously used items.

The similarity of items is determined by measuring the similarity in their properties. So, in this type of filtering method, there is no dependency on the rating records of other users in order to generate preferences for current users. Content-based systems focus on the properties of items. For example, if the user has purchased a book on Amazon.com which uses a recommendation system, then the user starts getting additional preferences for buying books from online book store which includes the same or similar keywords information for books.

Therefore, CBF-based recommendation systems generate recommendations using comparative representations of content relating an item to representations of content that are interesting to the user. In this method, news recommendation is done based on the content similarity between the news articles that the user has read, with the newly published news by considering news recent times. As such, many methods have classified this issue as an information retrieval (IR) issue, where content related to the user’s preferred choices is considered an enquiry, and unrated documents are evaluated on the basis of relevance/similarity to this enquiry [13,14,15]. In order to recommend one particular news to a user, the content-based approaches will get the previously rated news based on their category for example, (sports, science & technology, health, business) and then the news with the highest similarity to user preferences are recommended.

2.2.2. Collaborative Filtering (CF)

The other technique used in our study is collaborative filtering. In Collaborative filtering-based RS, the user will be suggested items that people with similar interests and preferences liked in the past. In a CF recommendation system, in order to suggest items to the user, the collaborative filtering recommendation system looks for the peers of the user, that is, a set of users that have similar interests in the item. Then, only the items that are the most liked by the peers of the user would be suggested [4]. So, collaborative filtering is a technology that aims to learn user preferences and make recommendations based on user and community data. It is a complementary technology to content-based filtering (e.g., keyword-based searching). Probably the most well-known use of collaborative filtering has been by Amazon.com for example where a user’s past shopping history is used to make recommendations for new products. However, various approaches to collaborative filtering have been proposed in the past in the research community.

In addition, Collaborative filtering (CF), as a kind of personalized recommendation technique, has been widely used in many domains [16,17]. However, collaborative filtering also suffers from a few issues, for instance, cold start problems, data sparsity, scalability and so on. These problems seriously reduce the user experience. Collaborative filtering recommends items to users according to their preferences. Therefore, a database of user history must be available. However, the database is always very sparse; that is, the user only rates a small number of items. Up until now, there have been many researchers who have focused on prediction accuracy and proposed some solutions.

CF has been becoming popular in both the academy and industry fields with great speed. Despite the overall success of CF systems, they suffer one serious limitation, namely the cold-start problem [4]. This cold-start problem includes two major aspects: new user and new item. Before a recommender system can present a user with reliable recommendations, it should know about this user’s preferences/interests, most likely from a sufficient number of behaviour records, e.g., ratings or log archives [19]. Since collaborative Filtering uses the similarity of users with respect to the items, they rated it is based on both user and item. Collaborative filtering is a technique that uses user-based similarity or item-based similarity. User-based Collaborative Filtering (UBCF) is one of the widely used recommendation technologies, remaining due to its convincing simplicity and quality of recommendations. It starts with the assumption that if a group of users have similar interests in their past, they will have similar interests in other items in the future [18]. The basic idea of User-Based Collaborative Filtering is to find a group of users who have a history of agreeing with an active user (i.e., they either gave similar ratings). Once a neighbourhood of users is identified, feelings from these neighbours are combined to produce recommendations for the active user. Item-based Collaborative Filtering (IBCF) is the similarity between items that are decided by looking at how other users have rated them. And it only considers users who have rated both items for each user and it calculates the difference in ratings for the two items finally, it takes the average of this difference over the users. For example, news in a similar category rated by different readers should be considered as similar items.

2.2.3. Hybrid Approach

In order to exploit the strengths of content-based and collaborative recommendations, one simple method is to accommodate both methods to obtain the two separately-rated lists of recommendations, and then arrive at a final list, which is a merger of the two results. The two predictions employing an adaptive weighted average are combined, where the weight of the collaborative component increases in line with the increase in the number of users accessing an item [20,21]. In this study, a hybrid approach is used which contains both collaborative filtering and content-based techniques.

The single recommendation system approach is not efficient enough to generate relevant and accurate recommendation preferences. So, hybrid recommendation systems came into existence to overcome the limitations of traditional recommendation approaches. The reason why the hybrid approach was selected is that when these two techniques work together it will increase the performance of the system and enhance the capability to answer user’s interest. These systems are based upon combining advantages of more than one traditional approach for recommendation generation for example; collaborative filtering approach with a content-based approach or collaborative filtering with demographic characteristics-based recommendation approach [5].

Figure 1. Recommendation System Taxonomies.

Figure 1. Recommendation System Taxonomies.

3. Methodology

3.5. Algorithm of the proposed model

The algorithm of our proposed system which represent all the approaches we used to overcome the problem formulated in our work is as follows. The pseudo-code consists of each of the separated approaches algorithm and their combination or the hybrid we developed in this study.

4. Result and Discussion

4.1. Results

We used 160 active users which contains different professions, both sex and different age groups. These selected users are the users who rated 32,624 news articles and make 303,594 rating data. Based on this data, the system generates the news for the new user registered and its categories the user into the appropriate category and if the category of the new user is not available in the system it creates a new category and adds the data to it. When a new user registered to the system and asks for a recommendation the system needs the user to request through the user interface prepared for the new user and use the information collected from user’s and articles recommendation is done on the interfaces.

To evaluate the performance, we depend on the objectives to recommend the more related or interesting news articles for new users we should have to evaluate the relatedness of the news with users. So, the popular and the most used metrics for any information retrieval to measure the relatedness or the interests are precision, recall and F-score. The following figure shows the descriptions of all metrics with their relationships [34].

Table 2. Evaluation Metrics [34].

Table 2. Evaluation Metrics [34].

	Recommended	Not Recommended
Good Articles	TP (True-Positive)	FN (False-Negative)
Not Good Articles	FP (False-Positive)	TN (True-Negative)

Figure 3. User Interfaces.

Figure 3. User Interfaces.

$$\mathit{P}\mathit{r}\mathit{e}\mathit{c}\mathit{i}\mathit{s}\mathit{i}\mathit{o}\mathit{n}=\frac{\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}\mathit{R}\mathit{e}\mathit{c}\mathit{o}\mathit{m}\mathit{m}\mathit{e}\mathit{n}\mathit{d}\mathit{e}\mathit{d}}{\mathit{A}\mathit{l}\mathit{l}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}\mathit{R}\mathit{e}\mathit{c}\mathit{o}\mathit{m}\mathit{m}\mathit{e}\mathit{n}\mathit{d}\mathit{e}\mathit{d}}$$

(1)

$$\mathrm{Or}Precision=\frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P}}$$

$$\mathit{R}\mathit{e}\mathit{c}\mathit{a}\mathit{l}\mathit{l}=\frac{\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}\mathit{R}\mathit{e}\mathit{c}\mathit{o}\mathit{m}\mathit{m}\mathit{e}\mathit{n}\mathit{d}\mathit{e}\mathit{d}}{\mathit{A}\mathit{l}\mathit{l}\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}}$$

(2)

$$\mathrm{Or}Recall=\frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{N}}$$

$$\mathit{F}-\mathit{S}\mathit{c}\mathit{o}\mathit{r}\mathit{e}=\frac{2\left(\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}\mathit{R}\mathit{e}\mathit{c}\mathit{o}\mathit{m}\mathit{m}\mathit{e}\mathit{n}\mathit{d}\mathit{e}\mathit{d}\right)}{2\left(\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}\mathit{R}\mathit{e}\mathit{c}\mathit{o}\mathit{m}\mathit{m}\mathit{e}\mathit{n}\mathit{d}\mathit{e}\mathit{d}\right)+\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}+\mathit{N}\mathit{o}\mathit{t}\mathit{G}\mathit{o}\mathit{o}\mathit{d}\mathit{A}\mathit{r}\mathit{t}\mathit{i}\mathit{c}\mathit{l}\mathit{e}\mathit{s}}$$

(3)

$$\mathrm{Or}F1-Score=\frac{2\mathrm{T}\mathrm{P}}{2\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P}+\mathrm{F}\mathrm{N}}$$

We have done the performance evaluation of our model in two ways. One is the experimentation to comparees each user with each other and the other is the experimentation of the system to compare the performance of categories of users which consists of many users.

i.

Experimentation for individual user similarity

For the individual users we evaluated by taking 16 which means 10% of active user for testing dataset from dataset which contains 160 active users. Then we remove the news rated by these 16 users and we register each of these 16 users as a new user and we run our new model to recommend the news for the users. And then we compared the previous news rated by the user with these new results. Then, we calculated the Precision, Recall and F1-score values as the formula we discussed in the previous section. According to this experimentation, results of the work is explained in the Table 3.

From the evaluation results, we have the accuracy of the news to each individual user with the values of 68.05% of average Precision, 42.46% of average recall and 52.1% of average of F1_score values as shown on Figure 4.

ii.

Experimentation by user category-based similarity

The other way we evaluated our performance is based on the user category. We took 5 user groups which of 10% of user groups out of user categories we have in our dataset. Then, we recommend some of new users similar to the category selected and we compare the accuracy performance by comparing the actual recommendation with the recommended for that category. According to this experimentation, results of the work is explained is shown in Table 4.

Based on this experiment we have the accuracy values of that performs the average precision of 93.75%, average recall of 40.25% and average F1-score of 56.31% and the average of the precision on this experiment and it is shown on Figure 5.

5. Conclusions

6. Future Work

References

1. C. Shahabi and Y. Chen, “Web Information Personalization: Challenges and Approaches,” in: Bianchi-Berthouze N. (eds) Databases in Networked Information Systems, vol. 95, pp. 1–10, 2003.
2. Bobadilla, J.; Ortega, F.; Hernando, A.; Gutiérrez, A. Recommender systems survey. Knowl. Based Syst. 2013, 46, 109–132. [Google Scholar] [CrossRef]
3. Lu, J.; Wu, D.; Mao, M.; Wang, W.; Zhang, G. Recommender system application developments: A survey. Decis. Support Syst. 2015, 74, 12–32. [Google Scholar] [CrossRef]
4. Adomavicious, G.; Tuzhilin, A. Towards the Next Generations of Recommender Systems: A Survey of the State-of-the-Art and Possible Extension. IEEE Trans. Knowl. Data Eng. 2005, 17, 734–749. [Google Scholar] [CrossRef]
5. Tranos Zuva, Sunday O. Ojo, Seleman M. Ngwira, and Keneilwe Zuva, “A Survey of Recommender Systems Techniques, Challenges and Evaluation Metrics. Int. J. Emerg. Technol. Adv. 2012, 2.
6. Available online: https://en.wikipedia.org/wiki/News.
7. Available online: https://en.wikipedia.org/wiki/Online_newspaper.
8. Mansi Sood, Harmeet Kaur, “Survey on News Recommendation,” International Journal of Advanced Research in Electrical Electronics and Instrumentation Engineering, Vol. 3, Issue 6, 2014.
9. F. Garcin, C. Dimitrakakis, and B. Faltings, “Personalized news recommendation with context trees,” in: The Proceedings of the 7th ACM conference on Recommender Systems, pp. 105-112, 2013.
10. Son, L.H. Dealing with the new user cold-start problem in recommender systems: A comparative review. Inf. Syst. 2016, 58, 87–104. [Google Scholar] [CrossRef]
11. P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom, and J. Riedl, “Grouplens: an open architecture for collaborative filtering of netnews”, CSCW, ACM, pp 175–186, 1994.
12. K. G. Saranya and G. S. Sadhasivam, “A personalized online news recommendation system,” International Journal of Computer Applications, pp 6–14, 2012.
13. M. Balabanović and Y. Shoham, “Fab: Content-based, collaborative recommendation,” Communications of the ACM, vol. 40, pp. 66-72, 1997.
14. Zheng, L.; Li, L.; Hong, W.; Li, T. PENETRATE: Personalized news recommendation using ensemble hierarchical clustering. Expert Syst. Appl. 2013, 40, 2127–2136. [Google Scholar] [CrossRef]
15. L. Li, L. Zheng, and T. Li, “Logo: A long-short user interest integration in personalized news recommendation,” in: Proc. the fifth ACM Conference on Recommender Systems, pp. 317-320, 2011. [CrossRef]
16. N. Zheng, L. Qiudan, L. Shengcai, Z. Leiming, “Which photo groups should I choose? A comparative study of recommendation algorithms in Flickr,” Journal of Information Science, vol. 36 no. 6, pp. 732–750, 2010.
17. Brynjolfsson, E.; Hu, Y. (.; Smith, M.D. Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Booksellers. Manag. Sci. 2003, 49, 1580–1596. [Google Scholar] [CrossRef]
18. Goldberg, D.; Nichols, D.; Oki, B.M.; Terry, D. Using collaborative filtering to weave an information tapestry. Commun. ACM 1992, 35, 61–70. [Google Scholar] [CrossRef]
19. Hofmann, T. Latent semantic models for collaborative filtering. ACM Trans. Inf. Syst. 2004, 22, 89–115. [Google Scholar] [CrossRef]
20. P. Cotter and B. Smyth, “Ptv: Intelligent personalized tv guides,” in: AAAI/IAAI, pp. 957-964, 2000.
21. M. Claypool, A. Gokhale, T. Miranda, P. Murnikov, D. Netes and M. Sartin, “Combining content-based and collaborative filters in an online newspaper,” in: Proc. ACM SIGIR Workshop on Recommender Systems, 1999.
22. L. Li, D. Wang, T. Li, D. Knox, and B. Padmanabhan, “Scene: A scalable two-stage personalized news recommendation system,” in: Proc. The 34th International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 125-134, 2011.
23. Urszula Kużelewska “Clustering Algorithms in Hybrid Recommender System on MovieLens Data,” Studies in Logic, Grammar and Rhetoric, vol. 37, no. 50, pp. 125–139, 2014.
24. L. Rokach and O. Maimon, “clustering methods,” Data mining and knowledge discovery handbook, ISBN. 978-0-387-24435, pp. 321-35, 2005.
25. Fraley, C.; Raftery, A.E. How Many Clusters? Which Clustering Method? Answers Via Model-Based Cluster Analysis. Comput. J. 1998, 41, 578–588. [Google Scholar] [CrossRef]
26. Marko Tkal c., Matev z. Kunaver, Andrej Ko. Sir, and Jurij Tasi c., “Addressing the New User Problem with a Personality Based User Similarity Measure,” no. 1, 2010.
27. Park, S.T., Pennock, D. M., Madani, O., Good, N., and DeCoste, D., “Naive filter bots for robust cold-start recommendations,” in: Proceedings of KDD ACM, vol. 06, 2006.
28. Salter, J.; Antonopoulos, N. CinemaScreen Recommender Agent: Combining Collaborative and Content-Based Filtering. IEEE Intell. Syst. 2006, 21, 35–41. [Google Scholar] [CrossRef]
29. Schein, A. I., Popescul, A., Ungar, L. H., and Pennock, D. M., “Methods and metrics for cold-start recommendations,” in: Proceedings of SIGIR ACM, vol. 02, pp. 253-260, 2002.
30. Pazzani, M., “A framework for Collaborative, Content Based and Demographic Filtering,” Artificial Intelligence Review, pp. 393-408, 1999.
31. Z. Lu, Z. Dou, J. Lian, X. Xie, and Q. Yang, “Content-based Collaborative Filtering for News Topic Recommendation,” in: Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI'15), pp. 217-223, 2015.
32. Hee-Geun Yoon, Hyun-Je Song, Seong-Bae Park, and Kweon Yang Kim, “A Personalized News Recommendation using User Location and News Contents,” Applied Mathematics & Information Sciences, vol. 9, No. 2, 2015.
33. Darvishy, H. Ibrahim, A. Mustapha, and F. Sidi, “New Attributes for Neighbourhood-based Collaborative Filtering in News Recommendation,” Journal of Emerging Technologies in Web Intelligence, vol. 7, no. 1, pp. 13–19, 2015.
34. F. O. Isinkaye, Y. O. Folajimi, B. A. Ojokoh, “Recommendation systems: Principles, Methods and Evaluation,” Egyptian Informatics Journal, vol. 16, pp. 261-273, 2015.
35. M. Sunitha and T. Adilakshmi, “Recommender Systems to Address New User Cold-Start Problem with User Side Information,” IOSR Journal of Computer Engineering (IOSR-JCE), vol. 18, no. 2, pp. 17–23, 2016.
36. Wai-Ho Au, Keith C. C. Chan, Andrew K. C. Wong, and Yang Wang, “Attribute Clustering for Grouping, Selection, and Classification of Gene Expression Data,” Technical Report, 2005.