05-11-2016, 09:15 AM
[attachment=74590]
Abstract : This paper is introducing the design of personalized research paper recommender system: Research Buddy. It provides personalized touch by creating separate user profiles. It uses this information to form clusters of researchers working in same field and having same interests. It uses fusion hybrid recommender to make recommendations. It suggests new research papers and maintains the history of research papers liked by the user in the past. It also takes care of long time and short time interest of the user while recommending.
Introduction:
The abundance of information available on the Web and in Digital Libraries, in combination with their dynamic and heterogeneous nature, has determined a rapidly increasing difficulty in finding what we want when we need it and in a manner which best meets our requirements.
Researchers have to spend many valuable hours searching for the information that they want. If these hours are spent on researching instead of digging the information then it can be boon to the research. This paper is introducing a design of personalized research paper recommender system: Research Buddy. It will recommend research papers to the researcher of his or her area of interest, hence saving time.
A survey was conducted amongst various researchers pursuing different fields to study the pattern in which they perform search on internet and the difficulties that they face. Following are the findings of the survey:
• It was found in the survey that mostly researchers are devoting 3 to 6 hrs daily for research out of which they have to spend 2 to 3 hrs locating the research paper of their interest.
• While searching some of the researchers judge the research paper through abstract and some prefer reading the whole paper.
• Many researchers agreed that sometimes the references given in the research paper are not relevant but mostly they are relevant.
• Another difficulty that they faced was that the search engine shows the same research papers every time.
• It was also found that researchers from the same lab often spend a considerable amount of time searching for published articles relevant to their current project. Despite having similar interests, they conduct independent, time consuming searches. While they may share the results afterwards, they are unable to leverage previous search results during the search process.
The proposed model Research Buddy tries to handle all these issues. It saves researcher’s valuable time by recommending research paper of his or her interest. It maintains user’s profile to provide personalization. It keeps record of people working in same field and provides this information on query.
Related Study
Although recommender systems are very popular in commercial applications these days, recommender systems for the academic research have also gained interest. This is noticed by the emergence of a lot of research papers about this topic presented in many conferences and journals. We describe some of the applications of research paper recommender systems.
In [8], a personalized academic research paper recommendation system is presented. It recommends relevant articles to the research field of the users. It is supposed that the users like their own articles. Based on this assumption, papers similar to the ones previously written by the system users are recommended as relevant to them. It uses text similarity to determine the similarity between two research papers and collaborative methods to recommend the items.
[18] designed a research paper recommendation system. It used content-based filtering technique as the recommendation technique. In this paper, Jaccard similarity coefficient is suggested to compute similarity between user’s query (user’s attributes) and the attributes of the papers. The recommendations suggested by the system were sent via email to the intended users.
Nascimento et al. provide another example of a content-based recommender system for scientific articles [14]. They point out that most of the recommender system approaches suppose that a large collection of scientific papers is available beforehand. It is true for some digital libraries like IEEE Xplore, but it does not hold for many other situations. Their proposed solution depends on publicly available scientific metadata. Instead of using user defined keywords, they generate keywords from a particular article that is presented by the user.
The hybrid approach of recommender systems was used for recommending research paper by [2]. Techlens recommends using different algorithms for recommending different kinds of papers. Various techniques of combining content-based filtering and collaborative filtering have been compared in this paper. It used dataset of CiteSeerX. Techlens also tries to point out that users with different levels of experience perceive recommendations differently.
In [17], another hybrid recommendation system is presented. Scienstein aimed to be powerful alternative to academic search engines. Instead of solely relying on text mining, it combines citation analysis, explicit ratings, implicit ratings, author analysis, and source analysis to a recommender system with a user-friendly GUI. CiteSeerX also uses citations to find similar scientific papers[22]. Some other applications with citation recommendation are presented in [19], [20], [21]. Albanian
In [1], content-based technique is adopted to recommend research papers. TF-IDF and cosine similarity were used to determine how relevant a research paper is to user’s query. It used Keyword-based Vector Space model to depict relationship between research paper and user’s query.
In [5], Personalized Research Paper Recommender System (PRPRS) is introduced. PRPRS UserProfile-based algorithm for extracting keyword by keyword extraction and keyword inference. It considers the title and text as an argument of keyword and execute the algorithm. Whenever collected research papers by topic are selected, a renewal of UserProfile increases frequency of each domain, topic and keyword. Each ratio of occurrence is recalculated and reflected in UserProfile. It uses Cosine Similarity to recommend initial paper for each topic in information retrieval.
In [3], a subspace clustering approach for recommender system is introduced. It studies the reading habits of other researchers who are interested in similar concepts. It uses collaborative filtering approach to collect data from other researchers browsing patterns, and avoids issues with the interpretation of content. It creates groups of people having similar interest. Such a group is represented by a subspace cluster. Finding these experts will ultimately help in finding research papers that form fields of interest.
In [23], Docear’s Research paper recommender system is being introduced. It manages the user’s data (papers, references, annotations, etc.) using mind maps. It allows all the users to create their own mind maps and then uses these mind maps to recommend the research papers. It uses content based filtering.
Proposed System
The proposed model “Research Buddy” aims to combine the already known concepts with new ones in order to create a holistic research paper recommender system. By combining different concepts, many disadvantages become obsolete. It tries to merge various techniques suggested earlier to create a personalized research paper recommender system that can overcome shortcomings of these techniques and provide user with a recommender system that can save their effort and time spent on locating research papers.
Methodology Research Buddy is composed of User Interface, Extractor, Cluster Manager, Cluster Profile, Profile Manager, User Profile, Fusion Hybrid Recommender and Monitor. It also uses CiteseerX Repository.
Collecting information about the users:
All the users who intend to use Research Buddy will need to get registered. User will be required to enter his or her details on first time logging into Research Buddy. This information will be used to create a separate user profile for the user in order to provide personalized touch. From time to time, this user profile will be updated.
The attributes that are required to store the information about the user are: user’s name, email-id, address, profession, area of interest, sub areas, research papers published by the user, and his preferences. A user profile will be created for the user using this information supplied by the user.
User profile thus created will reside on the local machine of the user in temporary folder. Whenever he or she logs in, then the user profile stored on his machine will be uploaded first and then based on this information recommendations will be made.
User profile will be updated whenever user accepts the recommendations or rates the recommendations received by him. Monitor will keep on monitoring user’s activities. Monitor will update the user’s profile based on explicit and implicit rating made by him.
Recommendations rated by him directly or accepted by him through implicit action will be added to user profile, so that it does not recommends the same research papers time and again.
Clustering the users:
After creating user profile, user will be added to one of the cluster having users working in the same field. If no such matching cluster exists then a new cluster will be created and allocated to him. Information about all the clusters is maintained in Cluster Profile. Cluster Profile will exist on the server hosting the Research Buddy.
Building the database:
Extractor extracts the research papers written by the user with the help of crawler. These extracted research papers are then converted into plain text by pdf to text converter. Header information which includes title of the research paper, author name, journal name, issn number etc. is then extracted from these text files. Citation count of this paper is handled by the citation checker. Indexing is being done of these extracted research papers. Indexed set of research papers will be handed over to collaborative filtering component.
Generating Recommendations:
Research Buddy uses Fusion Hybrid Recommender to make recommendations. It applies Collaborative filtering and Content-based filtering independently and then output of both is given as input to Fusion Hybrid Recommender. It merges the results of both the filtering algorithms together. The recommendations that are present in both are given higher ranking.
Content-based filter takes input from the user profile that is updated regularly. Collaborative filter takes its input from extractor and from cluster manager. Cluster manager handles all the clusters that are stored in cluster profile. Clustering helps the collaborative filter to get information about the research papers liked by other users present in the same cluster.
After generating recommendations, it matches them with the recommendations present in the user profile. It will filter out the already recommended research papers and will only pass on new recommendations.
Recommendations generated by Fusion Hybrid Recommender are further passed over to the user through the graphical user interface. It will show maximum ten recommendations at a time.
Analyzing the user’s satisfaction:
Users are asked to explicitly rate the recommendations generated by the recommender. It helps the recommender to judge the user’s preferences and hence improves accuracy of recommending the research papers. Explicit rating can be helpful in improving user profile as well as cluster profile.
Apart from explicit rating, Research Buddy monitors the actions of the user to judge implicit rating. This is the duty of monitor to analyze all the actions taken by the user. Monitor keeps record of all the actions taken by the user and based on this information it updates user profile as well as cluster profile.
Research Buddy: A research paper recommender system
Research Buddy is a unique recommender system which provides personalized touch by maintaining user profile. It applies both collaborative and content-based technique to recommend the research papers and keeps track of other users working in same area. Research Buddy helps the researcher by suggesting them research papers of their interest
User Profiling
In order to recommend papers to users we need to keep track of user’s interests. This can be done by maintaining user’s profile. User Profile represents the user’s tastes and opinions about the papers that are recommended to him. It can vary from a person to person. In order to give personalized touch user profiling is must.
User Profile can be helpful to find out long term interests and short term interests of the user. Long term interests can be judged by monitoring all the research papers that he has read and downloaded by now. Short term interests can be judged by the research papers recently read or downloaded. Short term interest can change from time to time. While recommending the research paper it is necessary to differentiate between these two. For example, a user may be interested in near past in image processing but now he is interested in semantic web. Recommender system should take care not to recommend research papers related to image processing now.
User profile can be updated by explicit and implicit rating. Research Buddy asks the users to select the research papers amongst the recommended research papers. These selected research papers are added to his profile. User can change area of interest at any time. This change is also noted in user profile so that next time research papers related to changed area of interest should be recommended. Research Buddy allows keeping more than one area of interests. Apart from this user is asked to explicitly rate the research papers.
Research Buddy implicit rating along with explicit rating. Sometimes user just selects a group of research papers which appears to him interesting but later on he may be ignoring many of these selected papers. It is duty of monitor to record all the actions taken by the user. Following user actions can be monitored which are enumerated in
Clustering
Research buddy uses clustering technique to group likeminded researchers. This technique is suggested in [3]. This approach is helpful to find groups of users who share a common interest in a particular field. This technique is quite helpful to collaborative filtering as it takes care of high dimensionality and scarcity, the two major problems faced by it.
Research Buddy is forming clusters based on the information present in user profiles. User profile is maintaining the record of area of interests of all the users and the research papers downloaded and liked by the users. Using this information it creates various clusters based on the area of interest. It then monitors the research papers of all the users who are present in same cluster. This helps to find out group of research papers that are liked by all the users.
Forming such kinds of clusters can be helpful to a researcher to know about other researchers working in the same field. One of the findings of our survey was that despite having similar interests, researchers belonging to even same department generally conduct independent, time consuming searches. This technique of Research Buddy will solve this problem.
Fusion Hybrid Recommender
The three basic approaches used in the design of recommendation systems are: collaborative filtering, content based and hybrid approach.
Collaborative Filtering
Collaborative filtering (CF) is one of the most successful techniques used in recommender systems. It has been used to recommend Usenet news[11], audio CDs[12], and research papers[13], among others. CF works by recommending items to people based on what other similar people have previously liked. CF creates neighborhoods of “similar” users (neighbors) for each user in the system and recommends an item to one user if her neighbors have rated it highly. CF is “domain independent” as it does not perform content analysis of item but rather relies on user opinions about the items to items to generate recommendation.
Content-Based Filtering
Systems implementing a content-based filtering (CBF) approach analyze a set of documents and/or descriptions of items previously rated by a user, and build a model or profile of user interests based on the features of the objects rated by that user.
Hybrid: Combining CF and CBF
Hybrid recommendation systems [7] usually use a combination of content-based and collaborative filtering recommendation for recommending items. This combined approach deals with the drawbacks of the above described ones, allowing for an initial content-based recommendation in cases of a cold start (lack of user profiles) [14]. The collaborative-filtering recommendation can improve the results by adding context-related information to the content-based approach.
Fusion hybrid recommender was suggested by [2]. It runs the collaborative recommender and content-based recommender in parallel and generates a final recommendation list by merging the results together. As mentioned above, in case of Research Buddy, collaborative recommender uses cluster profile as well as other research papers extracted from citeseerx repository and then makes the recommendations. Content-based recommender uses information available in user profile to make recommendations.
Fusion Hybrid recommender, first studies the recommendations by both the recommenders. It checks out the recommendations that are present in the both the lists and at what ranking. It sums up the rank of both the lists and terms it as score. Lower the score, higher will be its ranking in the final list. Recommendations that are not present in other list will be appended at the end of the final list.
Graphical User Interface of Research Buddy
To support the user in managing the information, Research Buddy provides user-friendly GUI. It tries to help the user in locating the desired research papers. Following are the various screens included in Research Buddy:
1. First screen Login screen :
This screen is used to enter username and password. If user is new, then new user’s sign in facility is included in it and forgot password and regeneration of password is being handled by it.
2. User’s Sign in Screen:
In this screen user will be asked to enter information required to build user profile. When user will enter his or her name then extractor will search for the research papers written by the user and will list all the research papers written by him. Provision to manually enter information related to research papers written by the user is also there.
Research Buddy judges the area of interests of the user through the list of research papers written by the user. Facility to manually enter it is also there. It accepts more than one area of interests. It asks the user to rank them. Through this information his user profile will be created. This user profile will be updated timely.
3. Welcome Screen
After signing in the welcome screen appears giving users different options like whether user wants to search for research paper or persons working in same area along with their demographic information.
4. Research Paper Recommender screen
This screen is divided into three parts. One part shows the name of the user and the area of interests selected by that user. User will be able to change the area of interest on this screen as well. Second part of screen shows the list of recommended research papers. It will be further divided into two parts. One showing new recommendations and the other showing the list of previously viewed research papers. This demarcation between old and new research papers will save the time of the user. Research papers which a user views or bookmarks or downloads will automatically be shifted into previously viewed research papers. Third part of the screen will be used for showing the abstract of selected paper.
Facility of narrowing of search is also provided i.e. if the user further wants to select sub area then it will be possible to do so. Another filtered list of recommendations will be provided. This feature can increase accuracy and satisfaction of user.
5. Explicit Rating screen
Before logging off user will be asked to rate the recommended papers. User will be asked to select the research papers which he wants to be added in his profile. This explicit rating of research paper will be helpful in learning the user’s preferences and interests.
Conclusion
This paper introduces the research paper recommender system: Research Buddy. The proposed model “Research Buddy” has combined already known concepts with new ones in order to create a holistic research paper recommender system. It uses fusion hybrid recommender to make recommendations, thus overcomes the shortcomings of both content-based and collaborative techniques. It maintains user’s profile, which helps in providing personalized experience to the user. It also maintains the information about various people working in the same field by making clusters. Suggested Graphical User Interface is also very powerful catering all the requirements of users and still is simple and easy to use.
This paper tries to explain each and every component of Research Buddy in detail. It will provide complete understanding of how Research Buddy works and handles various issues. In future, we plan to work on this design and make it actual.
Abstract : This paper is introducing the design of personalized research paper recommender system: Research Buddy. It provides personalized touch by creating separate user profiles. It uses this information to form clusters of researchers working in same field and having same interests. It uses fusion hybrid recommender to make recommendations. It suggests new research papers and maintains the history of research papers liked by the user in the past. It also takes care of long time and short time interest of the user while recommending.
Introduction:
The abundance of information available on the Web and in Digital Libraries, in combination with their dynamic and heterogeneous nature, has determined a rapidly increasing difficulty in finding what we want when we need it and in a manner which best meets our requirements.
Researchers have to spend many valuable hours searching for the information that they want. If these hours are spent on researching instead of digging the information then it can be boon to the research. This paper is introducing a design of personalized research paper recommender system: Research Buddy. It will recommend research papers to the researcher of his or her area of interest, hence saving time.
A survey was conducted amongst various researchers pursuing different fields to study the pattern in which they perform search on internet and the difficulties that they face. Following are the findings of the survey:
• It was found in the survey that mostly researchers are devoting 3 to 6 hrs daily for research out of which they have to spend 2 to 3 hrs locating the research paper of their interest.
• While searching some of the researchers judge the research paper through abstract and some prefer reading the whole paper.
• Many researchers agreed that sometimes the references given in the research paper are not relevant but mostly they are relevant.
• Another difficulty that they faced was that the search engine shows the same research papers every time.
• It was also found that researchers from the same lab often spend a considerable amount of time searching for published articles relevant to their current project. Despite having similar interests, they conduct independent, time consuming searches. While they may share the results afterwards, they are unable to leverage previous search results during the search process.
The proposed model Research Buddy tries to handle all these issues. It saves researcher’s valuable time by recommending research paper of his or her interest. It maintains user’s profile to provide personalization. It keeps record of people working in same field and provides this information on query.
Related Study
Although recommender systems are very popular in commercial applications these days, recommender systems for the academic research have also gained interest. This is noticed by the emergence of a lot of research papers about this topic presented in many conferences and journals. We describe some of the applications of research paper recommender systems.
In [8], a personalized academic research paper recommendation system is presented. It recommends relevant articles to the research field of the users. It is supposed that the users like their own articles. Based on this assumption, papers similar to the ones previously written by the system users are recommended as relevant to them. It uses text similarity to determine the similarity between two research papers and collaborative methods to recommend the items.
[18] designed a research paper recommendation system. It used content-based filtering technique as the recommendation technique. In this paper, Jaccard similarity coefficient is suggested to compute similarity between user’s query (user’s attributes) and the attributes of the papers. The recommendations suggested by the system were sent via email to the intended users.
Nascimento et al. provide another example of a content-based recommender system for scientific articles [14]. They point out that most of the recommender system approaches suppose that a large collection of scientific papers is available beforehand. It is true for some digital libraries like IEEE Xplore, but it does not hold for many other situations. Their proposed solution depends on publicly available scientific metadata. Instead of using user defined keywords, they generate keywords from a particular article that is presented by the user.
The hybrid approach of recommender systems was used for recommending research paper by [2]. Techlens recommends using different algorithms for recommending different kinds of papers. Various techniques of combining content-based filtering and collaborative filtering have been compared in this paper. It used dataset of CiteSeerX. Techlens also tries to point out that users with different levels of experience perceive recommendations differently.
In [17], another hybrid recommendation system is presented. Scienstein aimed to be powerful alternative to academic search engines. Instead of solely relying on text mining, it combines citation analysis, explicit ratings, implicit ratings, author analysis, and source analysis to a recommender system with a user-friendly GUI. CiteSeerX also uses citations to find similar scientific papers[22]. Some other applications with citation recommendation are presented in [19], [20], [21]. Albanian
In [1], content-based technique is adopted to recommend research papers. TF-IDF and cosine similarity were used to determine how relevant a research paper is to user’s query. It used Keyword-based Vector Space model to depict relationship between research paper and user’s query.
In [5], Personalized Research Paper Recommender System (PRPRS) is introduced. PRPRS UserProfile-based algorithm for extracting keyword by keyword extraction and keyword inference. It considers the title and text as an argument of keyword and execute the algorithm. Whenever collected research papers by topic are selected, a renewal of UserProfile increases frequency of each domain, topic and keyword. Each ratio of occurrence is recalculated and reflected in UserProfile. It uses Cosine Similarity to recommend initial paper for each topic in information retrieval.
In [3], a subspace clustering approach for recommender system is introduced. It studies the reading habits of other researchers who are interested in similar concepts. It uses collaborative filtering approach to collect data from other researchers browsing patterns, and avoids issues with the interpretation of content. It creates groups of people having similar interest. Such a group is represented by a subspace cluster. Finding these experts will ultimately help in finding research papers that form fields of interest.
In [23], Docear’s Research paper recommender system is being introduced. It manages the user’s data (papers, references, annotations, etc.) using mind maps. It allows all the users to create their own mind maps and then uses these mind maps to recommend the research papers. It uses content based filtering.
Proposed System
The proposed model “Research Buddy” aims to combine the already known concepts with new ones in order to create a holistic research paper recommender system. By combining different concepts, many disadvantages become obsolete. It tries to merge various techniques suggested earlier to create a personalized research paper recommender system that can overcome shortcomings of these techniques and provide user with a recommender system that can save their effort and time spent on locating research papers.
Methodology Research Buddy is composed of User Interface, Extractor, Cluster Manager, Cluster Profile, Profile Manager, User Profile, Fusion Hybrid Recommender and Monitor. It also uses CiteseerX Repository.
Collecting information about the users:
All the users who intend to use Research Buddy will need to get registered. User will be required to enter his or her details on first time logging into Research Buddy. This information will be used to create a separate user profile for the user in order to provide personalized touch. From time to time, this user profile will be updated.
The attributes that are required to store the information about the user are: user’s name, email-id, address, profession, area of interest, sub areas, research papers published by the user, and his preferences. A user profile will be created for the user using this information supplied by the user.
User profile thus created will reside on the local machine of the user in temporary folder. Whenever he or she logs in, then the user profile stored on his machine will be uploaded first and then based on this information recommendations will be made.
User profile will be updated whenever user accepts the recommendations or rates the recommendations received by him. Monitor will keep on monitoring user’s activities. Monitor will update the user’s profile based on explicit and implicit rating made by him.
Recommendations rated by him directly or accepted by him through implicit action will be added to user profile, so that it does not recommends the same research papers time and again.
Clustering the users:
After creating user profile, user will be added to one of the cluster having users working in the same field. If no such matching cluster exists then a new cluster will be created and allocated to him. Information about all the clusters is maintained in Cluster Profile. Cluster Profile will exist on the server hosting the Research Buddy.
Building the database:
Extractor extracts the research papers written by the user with the help of crawler. These extracted research papers are then converted into plain text by pdf to text converter. Header information which includes title of the research paper, author name, journal name, issn number etc. is then extracted from these text files. Citation count of this paper is handled by the citation checker. Indexing is being done of these extracted research papers. Indexed set of research papers will be handed over to collaborative filtering component.
Generating Recommendations:
Research Buddy uses Fusion Hybrid Recommender to make recommendations. It applies Collaborative filtering and Content-based filtering independently and then output of both is given as input to Fusion Hybrid Recommender. It merges the results of both the filtering algorithms together. The recommendations that are present in both are given higher ranking.
Content-based filter takes input from the user profile that is updated regularly. Collaborative filter takes its input from extractor and from cluster manager. Cluster manager handles all the clusters that are stored in cluster profile. Clustering helps the collaborative filter to get information about the research papers liked by other users present in the same cluster.
After generating recommendations, it matches them with the recommendations present in the user profile. It will filter out the already recommended research papers and will only pass on new recommendations.
Recommendations generated by Fusion Hybrid Recommender are further passed over to the user through the graphical user interface. It will show maximum ten recommendations at a time.
Analyzing the user’s satisfaction:
Users are asked to explicitly rate the recommendations generated by the recommender. It helps the recommender to judge the user’s preferences and hence improves accuracy of recommending the research papers. Explicit rating can be helpful in improving user profile as well as cluster profile.
Apart from explicit rating, Research Buddy monitors the actions of the user to judge implicit rating. This is the duty of monitor to analyze all the actions taken by the user. Monitor keeps record of all the actions taken by the user and based on this information it updates user profile as well as cluster profile.
Research Buddy: A research paper recommender system
Research Buddy is a unique recommender system which provides personalized touch by maintaining user profile. It applies both collaborative and content-based technique to recommend the research papers and keeps track of other users working in same area. Research Buddy helps the researcher by suggesting them research papers of their interest
User Profiling
In order to recommend papers to users we need to keep track of user’s interests. This can be done by maintaining user’s profile. User Profile represents the user’s tastes and opinions about the papers that are recommended to him. It can vary from a person to person. In order to give personalized touch user profiling is must.
User Profile can be helpful to find out long term interests and short term interests of the user. Long term interests can be judged by monitoring all the research papers that he has read and downloaded by now. Short term interests can be judged by the research papers recently read or downloaded. Short term interest can change from time to time. While recommending the research paper it is necessary to differentiate between these two. For example, a user may be interested in near past in image processing but now he is interested in semantic web. Recommender system should take care not to recommend research papers related to image processing now.
User profile can be updated by explicit and implicit rating. Research Buddy asks the users to select the research papers amongst the recommended research papers. These selected research papers are added to his profile. User can change area of interest at any time. This change is also noted in user profile so that next time research papers related to changed area of interest should be recommended. Research Buddy allows keeping more than one area of interests. Apart from this user is asked to explicitly rate the research papers.
Research Buddy implicit rating along with explicit rating. Sometimes user just selects a group of research papers which appears to him interesting but later on he may be ignoring many of these selected papers. It is duty of monitor to record all the actions taken by the user. Following user actions can be monitored which are enumerated in
Clustering
Research buddy uses clustering technique to group likeminded researchers. This technique is suggested in [3]. This approach is helpful to find groups of users who share a common interest in a particular field. This technique is quite helpful to collaborative filtering as it takes care of high dimensionality and scarcity, the two major problems faced by it.
Research Buddy is forming clusters based on the information present in user profiles. User profile is maintaining the record of area of interests of all the users and the research papers downloaded and liked by the users. Using this information it creates various clusters based on the area of interest. It then monitors the research papers of all the users who are present in same cluster. This helps to find out group of research papers that are liked by all the users.
Forming such kinds of clusters can be helpful to a researcher to know about other researchers working in the same field. One of the findings of our survey was that despite having similar interests, researchers belonging to even same department generally conduct independent, time consuming searches. This technique of Research Buddy will solve this problem.
Fusion Hybrid Recommender
The three basic approaches used in the design of recommendation systems are: collaborative filtering, content based and hybrid approach.
Collaborative Filtering
Collaborative filtering (CF) is one of the most successful techniques used in recommender systems. It has been used to recommend Usenet news[11], audio CDs[12], and research papers[13], among others. CF works by recommending items to people based on what other similar people have previously liked. CF creates neighborhoods of “similar” users (neighbors) for each user in the system and recommends an item to one user if her neighbors have rated it highly. CF is “domain independent” as it does not perform content analysis of item but rather relies on user opinions about the items to items to generate recommendation.
Content-Based Filtering
Systems implementing a content-based filtering (CBF) approach analyze a set of documents and/or descriptions of items previously rated by a user, and build a model or profile of user interests based on the features of the objects rated by that user.
Hybrid: Combining CF and CBF
Hybrid recommendation systems [7] usually use a combination of content-based and collaborative filtering recommendation for recommending items. This combined approach deals with the drawbacks of the above described ones, allowing for an initial content-based recommendation in cases of a cold start (lack of user profiles) [14]. The collaborative-filtering recommendation can improve the results by adding context-related information to the content-based approach.
Fusion hybrid recommender was suggested by [2]. It runs the collaborative recommender and content-based recommender in parallel and generates a final recommendation list by merging the results together. As mentioned above, in case of Research Buddy, collaborative recommender uses cluster profile as well as other research papers extracted from citeseerx repository and then makes the recommendations. Content-based recommender uses information available in user profile to make recommendations.
Fusion Hybrid recommender, first studies the recommendations by both the recommenders. It checks out the recommendations that are present in the both the lists and at what ranking. It sums up the rank of both the lists and terms it as score. Lower the score, higher will be its ranking in the final list. Recommendations that are not present in other list will be appended at the end of the final list.
Graphical User Interface of Research Buddy
To support the user in managing the information, Research Buddy provides user-friendly GUI. It tries to help the user in locating the desired research papers. Following are the various screens included in Research Buddy:
1. First screen Login screen :
This screen is used to enter username and password. If user is new, then new user’s sign in facility is included in it and forgot password and regeneration of password is being handled by it.
2. User’s Sign in Screen:
In this screen user will be asked to enter information required to build user profile. When user will enter his or her name then extractor will search for the research papers written by the user and will list all the research papers written by him. Provision to manually enter information related to research papers written by the user is also there.
Research Buddy judges the area of interests of the user through the list of research papers written by the user. Facility to manually enter it is also there. It accepts more than one area of interests. It asks the user to rank them. Through this information his user profile will be created. This user profile will be updated timely.
3. Welcome Screen
After signing in the welcome screen appears giving users different options like whether user wants to search for research paper or persons working in same area along with their demographic information.
4. Research Paper Recommender screen
This screen is divided into three parts. One part shows the name of the user and the area of interests selected by that user. User will be able to change the area of interest on this screen as well. Second part of screen shows the list of recommended research papers. It will be further divided into two parts. One showing new recommendations and the other showing the list of previously viewed research papers. This demarcation between old and new research papers will save the time of the user. Research papers which a user views or bookmarks or downloads will automatically be shifted into previously viewed research papers. Third part of the screen will be used for showing the abstract of selected paper.
Facility of narrowing of search is also provided i.e. if the user further wants to select sub area then it will be possible to do so. Another filtered list of recommendations will be provided. This feature can increase accuracy and satisfaction of user.
5. Explicit Rating screen
Before logging off user will be asked to rate the recommended papers. User will be asked to select the research papers which he wants to be added in his profile. This explicit rating of research paper will be helpful in learning the user’s preferences and interests.
Conclusion
This paper introduces the research paper recommender system: Research Buddy. The proposed model “Research Buddy” has combined already known concepts with new ones in order to create a holistic research paper recommender system. It uses fusion hybrid recommender to make recommendations, thus overcomes the shortcomings of both content-based and collaborative techniques. It maintains user’s profile, which helps in providing personalized experience to the user. It also maintains the information about various people working in the same field by making clusters. Suggested Graphical User Interface is also very powerful catering all the requirements of users and still is simple and easy to use.
This paper tries to explain each and every component of Research Buddy in detail. It will provide complete understanding of how Research Buddy works and handles various issues. In future, we plan to work on this design and make it actual.