Book Recommender Systems & Its Application
Recommender Systems have become inevitable today. In this era of fast changing trends providing effective recommendations to users makes you leap one step ahead of your competitors. It helps you provide a rich user experience and hence enhancing your user’s satisfaction. Book recommenders can be used to provide effective recommendations regarding which book a user might like to buy according to one’s interests. This paper gives an overview about the research that already exists on book recommender systems. It also briefs up about the algorithms used in book recommenders these days and their analysis on how effective they are.
KEYWORDS: Recommender Systems, Collaborative Filtering, Content-Based Filtering, Hybrid Systems, Association Rule Mining.
Nowadays there is a plethora of options available in every domain, but to select a specific option from such a diverse range of options is undoubtedly a challenging task for any person. This state of dilemma can be tackled by using recommender systems which can ease the user’s job by serving the recommendations for items which are more suitable to the user. “Recommender systems were originally defined as ones in which people provide recommendations as inputs, which the system then aggregates and directs to appropriate recipients.”
Book Recommender Systems are recommender systems for making recommendations regarding which book a user might like to buy. Just like when a customer goes to a store to buy some product, usually the salesman at the store shows recommendations to the customer based on the customer’s likes or dislikes observed by the clever salesman, here in digital world this task is accomplished by the recommender systems which produce pertinent recommendations for the user.
Recommender Systems can be used in several fields like for recommending goods on ecommerce websites, to recommend videos or movies, to recommend books or news articles, to recommend medicines in accordance with the symptoms a person has, to recommend the best possible route for your journey, tourism-based recommendations and several other domains.
In layman’s terms we may say that recommendation system’s applications are as good as the physical world salesman in the digital world. In general we can say it can almost be used in all those fields where one needs to grasp user’s interests to make useful recommendations. It is the digital world salesman. It is desirable to get good recommendations such that it easily catches hold of user’s interests and thus proving a boon in boosting sales. Book Recommender Systems work by gathering data about user’s likes or dislikes by collecting ratings given by users to books, books user has read, user’s profile etc.
This data is further processed by Book Recommender System to generate appropriate recommendations for books the user might like to read in future. There are various approaches used for processing this data like Collaborative Filtering, Content Based Filtering, Hybrid, Data Mining techniques etc.
Selecting the right book to read has always been a confusing task be it an amateur or a veteran in reading books. It is confusing to select a single book to read from so many promising choices, after all the world of books is no less than a fantasy land with a plethora of options. Book Recommender can ease this confusing process by providing suggestions about the books a user might like. Most algorithms for recommendation find a set of users whose rated and purchased items are similar to the user’s rated and purchased items. The algorithm gathers a list of such items and then removes the items that the user has already bought previously and recommends the rest to the user. 
In the paper “Book Recommendation Systems” , combination of different techniques have been used to make a book recommendation system. Techniques like collaborative filtering, content based approach, association rule mining have been used. In the paper “A Survey of Recommendation System: Research Challenges “, the challenges faced in different recommendation techniques have been surveyed, so that in future research can be done in those areas. According to it the primary goal of a recommender system is to predict ratings of non-rated user/items combination to provide apt recommendations. Challenges in recommender systems like sparsity problem, cold start problem, scalability, overspecialization problem have been discussed in the paper.
In the paper “A Novel Approach for Book Recommendation Systems”, the use of data mining techniques like Association Rule Mining has been discussed, also they have proposed a novel pattern mining algorithm called FP Intersect to overcome limitations posed by traditional Apriori. With many challenges like scalability issues, sparse rating matrices etc looming around it is surely a challenge for these systems to outwit the capabilities of a human brain in perceiving one’s interests implicitly.
Approaches used for Book Recommender Systems
In this technique the user-item interactions play a major role in recommending books. This technique mainly focuses on user-item relationships and their interdependencies to recommend books. Consider a user-item matrix as shown in diagram. Here items are books.
There are two approaches for collaborative filtering
i. User-Item based Collaborative Filtering
User based collaborative filtering is a method of personalized recommendations from a dataset, to a user based on similarity between the interest profile of a user and those of other users . Here users with similar choices are found based on the ratings they have given or their purchasing history. Then the items bought by these similar users are recommended to the user. As seen in diagram here u3 and u5 have both rated books Charlotte’s Web and Stephen Hawking, hence we can say these users are similar by calculating similarity between users using similarity measures like cosine similarity. Now u3 has also rated Harry Potter which u5 has not read so we can recommend Harry Potter book to u5. This is how User-Item based CF works.
ii. Item-Item based Collaborative Filtering:
Instead of matching users to similar customers, in item-item collaborative filtering we match each item bought by the user with similar items and then gather them into a recommendation list. To find most similar items a similar item table can be built by finding items customers purchase together.  As seen in figure item-item similarity can be computed for books which have been rated commonly. Similarity can be computed for books Wizard of Oz and Charlotte’s Web. Collaborative Filtering is a good approach for producing recommendations as it can produce novel recommendations for users. Limitations: But collaborative filtering approach also poses certain problems like cold-start problem, gray sheep user problem etc.
In the current recommendation systems some widely used similarity measures are cosine similarity measure and Pearson’s correlation coefficient measure.
Cosine Similarity: Also known as vector-based similarity, this formulation views two items and their ratings as vectors, and defines the similarity between them as the angle between these vectors:
Pearson’s Correlation Coefficient based Similarity: This similarity measure is based on how much ratings by common users for a pair of items deviate from average ratings for those items:
Content Based Filtering
Content Based Filtering is one of the commonly used techniques in recommendation systems. Pure Content Based Systems usually gather items which are similar to items user like based on analysis of item content . In this technique the attributes of users or items like user’s profile, keywords from books are considered for recommending various books to users. Content based recommendation system filter the entire set of books from the dataset based on the content of the book, where buyer is interested to buy. Recommendation system uses content based filtering for doing the separation and filtering of books from other books which is having similar kind of content. Also, this helps to discover the content of purchased history from the browsing data. This leads to good recommendation of books to the user based on their interest.
Typically, personalized profiles are created automatically through user feedback, and describe the type of books a user likes. In order to determine what items to recommend, collected user information is compared against content features of the books to examine. As shown below
System has a huge database consisting of the books to be recommended and the features of these books and it is termed as Book Profile here.
The users provide some sort of information about their preferences to the system. Combining the book’s information with the user preferences, the system builds a profile of the users.
According to the information existing in a target user’s profile, the system recommends suitable items to the user.
Limitations Content based filtering poses several limitations such as –Generally it cannot evaluate the quality of recommendation produced. Hence it cannot predict whether it is a good book with good overall ratings. It can only predict this book is similar to the book one liked.
Association Rule Based
In this technique various association rules among items (books here) are found to produce recommendations for user. Association rule mining finds interesting association and correlation relationship among large data set of items. Market Basket Analysis is the standard example for Association Rule Mining. Here customer’s purchasing trends are observed to decrypt interesting associations hidden in purchasing trends.
Different approaches for recommendation systems like collaborative filtering, content based filtering pose different limitations. The effect of these limitations can be reduced by using combined approach. So in hybrid systems we combine approaches like content based and collaborative to alleviate their respective limitations and generate better recommendations.
Book Recommendation System Based on Collaborative Filtering and Association Rule Mining for College Students. Anand Shanker Tewari, Kumari Priyanka
Amazon.com Recommendations Item-to-Item Collaborative Filtering Greg Linden, Brent Smith, and Jeremy York, Amazon.com
G.D. Linden, J.A. Jacobi, and E.A. Benson, Collaborative Recommendations Using Item-to-Item Similarity Mappings, US Patent 6,266,649 (to Amazon.com), Patent and Trademark Office, Washington, D.C., 2001.
Book Recommendation System through Content Based and Collaborative Filtering Method Ms. Praveena Mathew1
Cite this Essay
To export a reference to this article please select a referencing style below