Recommendation Systems and Methods of Student Investigation - Research Proposal Example

Student’s Name Instructor Course Date The Imрасt of Rесоmmеndаtiоn Systеms Оn Thе Меthоds of Studеnt Rеsеаrсh for Study 1. Introduction Over time, information that is useful to researchers is becoming more and more massive. This information is either in the form of structured or unstructured data and given its large volume, students and other researchers are faced with the difficulty of getting the right information within the least time possible. Additionally, using traditional software and database techniques to process such massive data is proving to be a hard task. The implication is that such large and fast changing data is said to be beyond the processing capacity for traditionally applied information processing techniques. Additionally, research students are faced with the problem of information overload which translates to difficulties in obtaining personalized recommendation, services and research content[Sch08]. Since their first appearance in the mid 1990s, recommendation systems have become an important area of study especially due to their abundance applications that enable users to handle information overload. Some of the application areas for recommendation systems include information and data mining in entertainment, electronic shops, or trends by people in social media. Recommendation systems are also significant to students in recommending books, CDs and other credible information sources like books, movies, news among other source thereby promoting technology enhanced learning[Man10]. Over the last few decades, there has been an ever increasing demand for web in commerce, governance, and learning. Additionally, there have been trends towards massification of interactive media which then led to an increasing need for more effective techniques for information filtering. While much focus has been directed towards development of recommendation systems to meet various personalized and customizable demands, this paper proposes a study on the impact of recommendation systems on students’ research method. In the past, research has focused on the impact of recommender systems on e-commerce domain. In e-commerce, recommendation systems use personalized navigation through large data amounts for product purchase advice and this impacts sale diversity. This is a clear indication that while much time is spent developing recommendation systems, little is done to study the impact of academic resources suppliers on the research students. Further, much research has been identified to focus on recommender system algorithms and how to improve primary technology such that they offer more robust recommendations[Nau10]. Apart from e-commerce, the use of recommender systems has been used in recommending content within social networks[Mic12] which emphasizes that the use recommender has commendable impacts to student learning. This paper compares the use of traditional logistics and linear regression techniques to prove that recommender systems are an effective remedy for improving students’ research methods. Additionally, Shani and Gunawardana [Sha13] reveal that the fact that recommender systems have gained increased use with Retailers such as Netflix, Amazon and Microsoft is enough proof that research students can use recommender systems for data mining to meet their research needs hence improve learning[Cas07]. 2. Aims and Objective For this study, the aim will be to establish the impact of recommendation systems on the methods of student research and to compare recommendation systems to traditional techniques such as logistic and linear regression. 3. Problem statement and justification Recommendation systems are an effective way of assisting users to deal with information overhead[Kan10]. Users are assisted through provision of guidance in a way that is personalized and directed to interesting resources. This personalized guidance and advice to users on products and services could be of interest especially in the use by student researchers. According to Mican, Mocean and Tomai [Mic12], recommender systems offer recommendations to users depending on information obtained from gathered user-generated collective intelligence. It is to this effect that these systems have gained increased application in content recommendation within social media. The result is that recommender systems are an everyday part of life within social media and networking recommendation. Initially, recommender systems have been ranked depending on their power of prediction which defines accuracy in prediction of user’s choices[Sha13]. However, today, the accuracy of prediction for any recommendation system is not sufficient to deploy them. This is the case given that recommendation systems’ accuracy of application is dependent on the context of given application. In this case, student research methods can be effectively applied to any recommendation system regardless of their interaction properties to provide commendable results[Sha13]. With defined research method properties, the recommendation system then compares the gathered data with other collected data that is or is not similar and uses comparison result to calculate the a recommendation list of the most accurate object for the user or researcher’s needs. For best recommendation list, Knijnenburg, Willemsam and Gantner [Kni12] suggest the chosen recommender system should manage to link effectively the objective system aspects to the objective user behavior. Additionally, the system must be in a position to incorporate influence of situational and personal characteristics of user experience. From Knijnenburg, Willemsam and Gantner [Kni12] the perceptions of recommendation quality and range are crucial determinants of predicting the impacts of objective system elements on the experience of the user. User experience components in this case include process in terms of perceived efforts, systems in terms of perceived effectiveness of the system, and outcomes in terms of choice satisfaction. Towards making the choice on the best for recommendation systems, the researcher can obtain the best category from collaborative, content-based and hybrid[Ado05]. 4. Literature review 4.1 Comparing recommender systems to logistic and linear regression Like recommendation systems, regression models are a function of data mining used to predict number. Some applications of regression methods include mortgage rates, house values, distance, and temperature amongst others. In regression, algorithms are used to estimate the value of the target as a function of the predicator. Through regression analysis, it is possible to determine the parameter values for a function that makes the function to best fit a set of data functions provided[Yan09]. However, in world where big data requires analysis for everyday decisions, regression models like Logistic and linear regression are not applicable especially given that they deal with continuous target as a function of predicator, a quantification of error and parameter sets[Con14]. Despite the drawback of being used only on continuous numerical target, regression models are useful tools in predicting the future, decision support, correcting errors, provision of new insights. One major advantage of using regression models is so as to predict the future through their use of research and analysis to predict what is likely to happen in the future. Future prediction is essential in ensuring that businesses improves where need be[Wan03]. In terms of supporting decisions, regression and forecasting ensure that business management is provided with sufficient information to make credible business decisions[And08]. Additionally, when regression and forecasting provide business management with new insights since it uncovers patterns and relationships that had previously not been noticed or considered. When using multi-variables, business managers test several independent variables which explain numerous characteristics of dependent variables[And08]. Being in a position to allow managers to test for several variables improves the accuracy of prediction. 4.2 Recommendation systems In a world where data is no longer linear and continuous but extremely massive, structured and unstructured, better methods are needed to analysis and decision making[Ric11]. Towards overcoming the drawbacks of regression, it is apparent that recommender systems are more effective and efficient techniques of assisting users to make the better choices[Gha10]. In particular, matrix factorization in recommender systems are designed such that they effectively and efficiently work with large contents and catalogs to provide users with personalized subsets of items customized to meet their own needs[RDG13]. Additionally when using factorization, researchers can effectively manage content overload while at the same time maneuvering into a pool of content resource discovery[Ric11]. The main contributing factor towards the recommending Recommender Systems for research students is that these systems are normally focused for a specific type of item. Some examples include Netflix systems for CDs, Amazon.com systems for books. Besides focusing on a given item, recommender systems also have designs, graphical user interface, and primary recommendation techniques that generate all customized recommendations that offer effective and useful suggestions for that specified item. According to Ricci, Lior and Shapira [Ric11], recommendation systems are also directed towards users with insufficient experience individually or persons with little competence to handle extremely overwhelming number of options offered by a given website. In particular sites like Amazon.com, Recommender systems enable users to select a specified book to read in Amazon.com given that the website personalizes the online store for each customer[Ric11]. The ability of recommender systems to personalize the online store in Amazon.com ensures that each user or group of users has diverse suggestions. In order to complement the personalized recommendations, recommender systems at amazon.com also offer non-personalized recommendations which are much simpler to generate unlike with personalized recommendations where the systems derives the user preferences from implicit or explicit feedback. Through explicit feedback, the user has the opportunity to rate items like is the case with YouTube[McN06]; critique items; assign weights to attributes items[Hau04]; or indicate specified needs as is the case with HP.com[Fel01]. Conversely, implicit feedback recommenders evaluate purchasing or clicking behavior like in Amazon.com[Hau09]. In the end, the calculation of recommendations by the system is done on the basis of comparing the preferences by users to those of the catalog items or content-oriented recommender systems or users’ preferences are compared to other user’s preferences through collaborative filtering systems[Kni121]. For researchers, collaborative filtering offers a good way for a research student to rely on what their peers recommend when selecting a book to read. Like student, recruiters also make their recruiting decisions based on recommendation letters. 4.3 Survey of recommender systems The evolution of recommendation systems traces its roots back to the far-reaching work in forecasting theories; approximation theory; management science; and information retrieval. Recommendation systems are also linked to consumer choice modeling as used in marketing and emerged as an independent area of research in the mid-1990s[Man10]. During the mid-1990s, recommendation systems emerged in response to researchers increased focus on the recommendation problems that entirely called for rating structures to make decisions. The combination of all these areas was meant to minimize the problems of ratings estimations on items that a user had not seen but which other similar users had. Insightfully, the estimation rating was based on the rating given by the user and the rating given by other information. After estimation was calculated, it could be used for those items that were not yet rated. The unrated item with the highest estimated rating could then be recommended to the user. The result was the formulation of content-based recommender systems and collaborative filtration recommendations. Hybrid recommender systems were later developed to combine both collaborative filtering and content based recommender systems[Gha10]. 4.4 Recommender system strategies and matrix filtration 4.4.1 Content based filtering According to Koren, Robert and Volinsky [Kor09], recommender systems are based on one or two strategies. Content-based filtering strategy ensures that each user or product has a unique profile that characterizes them. Although this strategy may not be so much used by student researchers, it is used by they are useful in areas like music genome project as used for the internet radio service Pandora.com. With the profile, it is then possible to have the user or product to link users to matching products. This means that, in order to use content based recommender systems, there has to be external information gathered and made available to user with profile. Such information may not be easy to locate or collect and this forms the drawback to content-based recommendation. 4.4.2 Collaborative filtering Collaborative filtering efficiently overcomes the barrier of content-based filtering by having no domain associations at all. Recommendations through this approach are merely founded on the basis of analyzing what the relationships existing between the users and interdependencies among products. The result is new user-item[Hau08] association that is obtained by addressing data elements that are hard and elusive to profile using content based filtering. According to Koren, Robert and Volinsky [Kor09], the two main approaches to collaborative filtering that could extremely benefit research students are neighborhood methods and Latent factor models. In the neighborhood collaborative filtering methods, the major aspect is to compute the associations between users or items. In item based approach, the preference of a user for an item is founded on the available ratings for neighboring items using the user’s sense[Her04]. When a single user rates similar items products using the same rating, then these products are neighbors. With latent factor model, ratings are explained through the characterization of both the item and users on factors inferred from the patterns of rating. 4.4.3 Matrix factorization Matrix factorization is utilized mostly in collaborative filtering and most successfully in latent factor models[Kor09]. Basically, matrix factorization involves the characterization of both the user and the items by the use of vector attributes attained from item rating patterns that have been inferred. High correspondence between user and item attributes results to a vector defined by a recommendation[Ren10]. Collaborative filtering and matrix factorization have gained popularity over the recent years given their efficient combination of accurate predication and good scalability. Additionally, the approaches offer flexibility in that it is possible for student researchers or users to model various real-life situations. Utilization of collaborative filtering and matrix factorization techniques is useful to researchers especially in patent pending such that the users’ tastes and preferences can be learnt and the quality of the recommendation constantly enhanced[Bob09]. In order to provide best objects to a researcher or user, collaborative filtering with matrix factorization systems compares the gathered data with that collected from others, whether similar or not, to come up with a recommendation list[Wel06]. 5. Methodology This section addresses the manner in which this research will be carried out and the methods to be used in information gathering. The section also provides the study limitations. Finally, there is justification to the method used in data collection. 5.1 Research method The study will utilize qualitative research methods for data collection and analysis. Qualitative research samples extensive literature which is then reviewed for its take on the impact of recommender methodologies on research student methods. Literature will be obtained from numerous sources which will then be evaluated and categorized on depending on the theme presented. Categorization themes will focus the function arising as a result of using recommendation systems for their research. 5.2 Limitation Since recommender systems use users’ opinions concerning particular items within a given domain to assist people in their choice of other items, such assistance to users on selecting an item can result to influencing the opinion of the users’ on the items[Cos03]. The implication is that the influenced opinions no longer hold high value in recommendations delivery for other users. Additionally, there are manipulators with the tendency of making their systems generate manmade high or low recommendations and they might benefit if their efforts end up influencing users to change their opinion. The impact of the recommender systems on research student methods will be limited to the assumption that the recommendation lists issued to researcher have negligible influence from system interfaces and predication display during the time the researchers rated the books. 5.3 Data analysis The qualitative data will be analyzed through an extensive review of the literature. This will help in understanding what others have done concerning in areas related to the research. This will give me as researcher insight and knowledge required for making necessary decisions during data collection. Some themes being sought after during research include researcher finds some good books; find all good books; given annotation in context; a sequence of books is recommended; a bundle of books is presented; identification of credible recommender; user given platform to express self; and provided with opportunity to help and influence others. The results are intended to show that recommendation systems predications influence the research students’ method by confirming their opinions or changing their opinions depending on what majority researchers in the same field prefer. Additionally, sometimes the users may detect the errors arising from deliberate or unintentional errors and end up retaining their opinion about a given product. However in most cases, most students are not quick to note such errors which are why the research will take an assumption that such errors if any are minimal and have no strong impact on the findings. 6. Plan and Timescale Conclusion The main aim of the study will be to establish the impact of recommendation systems on the methods of student research and to compare recommendation systems to other traditional techniques such as logistic and linear regression. From the study, it is expected that, regardless of the existing limitations, recommendation systems are efficient and effective alternatives to search algorithms. In a time and age where information searching from online sources is too massive for traditional regression techniques, recommender systems have proven to be a useful tool in providing users with personalized recommendations. From the successful application of recommendation systems in social networking and in sales diversity, it is clear those information recommendation systems through collaborative filtering and matrix factorization present an essential data mining process especially for researchers. Works Cited Sch08: , (Schiaffino, Garcia and Amandi 1745), Man10: , (Manouselis, Drachsler and Vuorikari 2812), Nau10: , (Naughton and Lin 1-5), Mic12: , (Mican, Mocean and Tomai 2), Sha13: , (2), Cas07: , (Castro, Vellido and Nebot 136), Kan10: , (Kantor, Ricci and Rokach 8), Mic12: , (1), Sha13: , (Shani and Gunawardana 2), Kni12: , (1), Ado05: , (Adomavicius and Tuzhilin 735), Yan09: , (Yan, Su and Scientific 25), Con14: , (Oracle Corporation), Wan03: , (Wang and Jain 196), And08: , (Anderson 722), Ric11: , (Ricci, Lior and Shapira 1-2), Gha10: , (Ghauth and Abdullah 1042-1045), RDG13: , (Gravity R&D), Ric11: , (Ricci, Lior and Shapira 1-5), Ric11: , (1), Ric11: , (Ricci, Lior and Shapira 13), McN06: , (McNee, Reidl and Konstan 1097), Hau04: , (Haubl, Dellaert and Murray 207), Fel01: , (Felix, Niederberger and Steiger 399), Hau09: , (Hauser, Urban and Liberali 202), Kni121: , (Knijnenburg, Willemsen and Gantner 442), Man10: , (Manouselis, Drachsler and Vuorikari 2815), Gha10: , (Ghauth and Abdullah 1044), Kor09: , (42), Hau08: , (Hauger, Karen and Schimdt-Thieme 525), Kor09: , (43), Her04: , (Herlocker, Konstan and Terveen 5), Kor09: , (Koren, Robert and Volinsky 48; Rendle, Freudenthaler and Schimidt-Thieme 7), Ren10: , (Rendle, Freudenthaler and Schimidt-Thieme 8), Bob09: , (Bobadilla, Serradilla and Hernondo 261), Wel06: , (Wells), Cos03: , (Cosley, Lam and Albert 1), Read More