|
Xuetao Guo, Faculty of Information Technology [HREF1], University of Technology, Sydney [HREF2], PO Box 123, Broadway, NSW 2007, Australia. xguo@it.uts.edu.au
Jie Lu, Faculty of Information Technology [HREF1], University of Technology, Sydney [HREF2], PO Box 123, Broadway, NSW 2007, Australia. jielu@it.uts.edu.au
Simeon Simoff, Faculty of Information Technology [HREF1], University of Technology, Sydney [HREF2], PO Box 123, Broadway, NSW 2007, Australia. simeon@it.uts.edu.au
Many E-commerce websites attempt to develop personalized features to encourage users' repetitive visits. Yet, there is less attention about the applications of personalization technologies in E-government services. In this study, we present a classification of personalization techniques. Also, a novel recommendation approach is proposed to improve the existing techniques by the integration of user-based and item-based collaborative filtering recommendation techniques. A recommender system prototype, named Smart Trade Exhibitions Finder, is developed to help companies choosing the right trade exhibitions. The outcome of this study will have tremendous significance in overcoming the drawback of existing recommendation approaches.
The continued explosion in the amount of web content and information are making the need of effective personalized online information delivery more acute. Many websites attempt to develop personalization features to encourage users' repeat visits. IDC ([HREF3]) forecasts that there would be 212% annual growth in personalization applications over the next decade [1]. With personalized services and products are becoming more common on the Internet, the interest on prsonalization technologies is accordingly growing. Personalization is concerned with building a closer relationship and understanding the needs of individuals or group of customers, whether on the Internet or not. Personalization is experiencing widespread adoption in application areas such as customer relationship management, E-commerce interaction and intimacy, and employee development and training. Also, as one of the most popular application of personalization techniques, recommender systems have gained much attention in the past ten years. Recommender systems aim at filtering out the uninterested items (or predicting the interested ones) automatically on behalf of the users according to their personal preferences. A recommender system helps users to either predict whether a particular user will like a particular item, or to identify a set of N items that will be of interest to a certain user [2].
Meanwhile, the government is committed to maximize the opportunities provided by the Internet. The potential for government using Internet to enhance services to its citizens, businesses and the other agencies are now more evident than ever before. Thus, in the most of developed countries and some of developing countries, E-government applications are growing rapidly. E-government technology promises to make government more efficient, responsive, transparent, and legitimate [3]. The challenge for government is to continually embrace the opportunities that the Internet provides, and ensure that citizens, business and community needs and expectations are met. The amount of information available on the Internet is overwhelming, so that the users of E-government websites are constantly facing the problem of information overload. Obviously, the difficulties of locating right information would increasingly affect the loyalty on returning to those E-government websites. Fortunately, personalization and recommender systems can provide a solution for this problem.
Recently, the fifth annual Accenture eGovernment study [4] indicates that personalization in E-government is emerging. The personalization of E-government services can be seen as an evolution of the intentions-based approach. The Australian Trade Commission (Austrade) is the Australian government agency that helps Australian companies developing international markets. Organizing trade exhibition participation and helping more Australian companies succeed in export are one of important tasks conducted by Austrade. International trade exhibitions are frequently used in exporting firmsĄ¯ marketing strategies and are of great value for exporting firms to communicate with potential and current customers from many countries in a short period of time ([HREF4]). As the number of trade exhibitions' visitor increase, trade exhibitions have become a relatively important promotion tool for industrial firms. It is vital that government help companies choosing right trade exhibitions. However, Austrade can only offer the simple database match function from their online services.
Therefore, in this study, a recommender system prototype, named Smart Trade Exhibitions Finder (STEF), is developed to help companies choosing the right trade exhibitions. Also, a novel recommendation approach is proposed to improve the existing recommendation techniques by the integration of user-based and item-based collaborative filtering. The outcome of this study will have tremendous significance in overcoming the drawback of existing recommendation approaches.
The rest of this paper is organized as follows. Section 2 presents the related works based on the previous literatures. Then, a classification of personalization is presented in Section 3. Next, In Section 4, we describe the architecture and implementation in the STEF. Finally, in Section 5, conclusions are drawn and future study problems are recommended.
Because personalization is relatively a new field, different authors have provided various definitions of the concept. According to Eirinaki and Vazirgiannis [5], web personalization refers to the process of customizing the content and structure of a website to the specific and individual needs of each user taking advantage of user's navigational behavior. Web personalization can sharply respond to a user's unique and particular needs. Mobasher et al. [6] defined web personalization as an act of response according to the individual user's interest. Web personalization was defined by Mulvenna et al. [7] as any action that adopts the information or services provided by a website to the needs of a particular user or a set of users, taking advantage of the knowledge gained from the usersĄ¯ navigational behaviour and individual interests, in combination with the content and structure of the website. In this study, we extend personalization as a process of gathering and storing information about site visitors, analysing the information, and delivering the right information to each visitor at the right time.
Many practitioners and researchers are investigating into various issues of personalization. The three conceptual layers [8] can describe the possible personalization of a web application. The first is personalized content. It means that each user can obtain a different content. The typical example of personalized content is the definition of the price of an article based on the user profile. The second is personalized navigation. This describes primitives by which users browse the content of web applications. Personalized navigation means to add, remove links, create a new collection or modify an existing one for a given user. A typical example of link addition is cross-selling. For instance, in the online bookshop Amazon ([HREF5]) website, the addition in a product page of links points to related items deemed potentially interesting for the customer. The third is personalized presentation, which defines the graphical resources of the pages and their layout adaptation with respect to user styles (congnitive, utilisation,navigation, etc).
On the other hand, various approaches for recommender systems have been developed. Previous research efforts show that the most existing recommender systems adopt two main types of techniques: content-based filtering (CBF) approach and collaborative filtering (CF) approach and their combinations. A CBF approach relies mainly on content and relevant profiles to offer recommendations. CF approach offers recommendation based on the similarity of groups users. In the content-based approach, it recommends web objects that are similar to what the user has been interested in the past. CF has been known to be the most popular recommendation technique that has been used in a number of different applications such as recommending web pages, movies, articles and products. CF-based approaches can be divided into two types: user-based and item-based CF. User-based CF is implemented in two steps: (1) a set of k-nearest neighbours (KNN) of target user are computed. This is performed by computing correlations or similarities between user records and the target user; (2) producing a prediction value for the target user on unrated (or unvisited) items. A major problem with this approach is the lack of scalability and sparsity. The complexity of computation increases linearly with the user and rating number. With millions of customers and products of real world situations, user-based approach can suffer serious scalability problems. In contrast, item-based CF is used to deal with the scalability problems in the traditional CF algorithms. Item-based CF avoids the bottleneck in user-user computations by first considering the relationships among items. Rather than finding user neighbours, the system attempts to find k similar items that are rated (or visited) by different users in some similar way. Then, for a target item, predictions can be generated, for example, by taking a weighted average of the target user's item ratings (or weights) on these neighbour items. Thus, these algorithms alleviate the scalability problem that exists in user-based CF algorithms. Item-based CF has been shown to achieve prediction accuracies that are comparable to or even better than user-based CF algorithms[8]. Item-based CF algorithms still suffer from the problems associated with data sparsity, and they still lack the ability to provide recommendations or predictions for new or recently added items.
Personalization has been adopted by the different techniques. Based on the literature review, we classify the existing personalization techniques into the following five categories:
3.1. Profile-based Personalization
In order to be able to purchase or receive advanced services from most websites, users are required to register and enter personal information (user profiles) such as gender, age, interests, etc. website store this information in a database on the web server. Websites also use user profiles for personalized services. The userĄ¯s postal code provides economic information so that the website can reorganize product access according to a customerĄ¯s economic profile. For example, a featured wine sale might not be advertised to users residing in an area known to have a depressed economy.
3.2. Link Personalization
This strategy involves selecting the links that are more relevant to the user, changing the original navigation space by reducing or improving the relationships between nodes. E-commerce applications use link personalization to recommend items based on the clients buying history or some categorization of clients based on ratings and opinions. Users who give similar ratings to similar objects are presumed to have similar tastes, so when a user seeks recommendations about products, the web site suggests the most popular or best correlated products for that class. Link personalization is widely used in Amazon ([HREF5]) to link the home page with recommendations, new releases, and shopping groups.
3.3. Content Personalization
We can say that content is personalized when nodes (pages) present different information to different users. The difference with link customization is subtle since when links are personalized, part of the contents (the link anchors) present different information. However, we will refer to content personalization when substantive information in a node is personalized, other than link anchors. Content personalization can be further classified into two types: node structure customization and node content customization.
3.4. Structure Personalization
Structure personalization usually appears in those sites that filter the information that is relevant for the user, showing only sections and details in which the user may be interested. The user may explicitly indicate his preferences, or it may be inferred (semi-) automatically from his profile or from his navigation activity.
3.5. Peer-to-peer Recommender Systems
Peer-to-peer recommender systems provide personalized recommendations based on usersĄ¯ preferences. Several recommender systems combine collaboration filtering and content filtering to improve recommendations [9].
Each of the existing recommendation techniques has strengths and weaknesses. Combining the different techniques to achieve peak performance is becoming a common thread in recommender system research. Thus, the recommender system STEF proposed in this study combines user-based and item-based CF methods. The STEF architecture is shown in Fig 1.
STEF recommender system constitutes of three components:
- Data Collector. It involves the collection of business information, preference, and trade exhibition information;
- DB Builder. It involves the integrated development of databases - business profile DB and trade exhibition DB;
- Recommendation Engine. It suggests trade exhibition to individual business according to the preferences specified by the business.
Our approach looks into the set of items and computers how similar the items are to target item i and then selects k most similar items {i1, i2, i3, Ąik}. The recommendation is generated through a set of algorithms. Our approach generates recommendations in the following steps:
-computing integrated user-based CF similarity,
-computing item-based CF similarity,
-integrating user-based with item-based CF similarity,
-generating recommendation.
In our case, from the application point of view users are the businesses and the items are the exhibitions.
As a data collect interface and test bed, the system is implemented in Microsoft IIS 6.0 environment with ASP programming language. The example of system interface is showed in Figure 2.
International trade exhibitions are a fantastic way to find new markets and customers for international businesses. There are thousands of trade exhibitions held annually around the world. The recommender system developed in this study can help businesses find the right ones for them, and therefore will reduce the time, cost and risk involved in selecting, entering and developing international markets. The study proposes a recommender system STEF to suggest relevant trade exhibitions to exporters for improving their product export and international businesses. A novel approach is used by the integration of user-based and item-based CF. The further study will address on the evaluation of its accuracy and performance, and the comparison of different recommendation approaches in the test bed. Last, but not least, we plan to evaluate the system usability.
[1]Amoroso, D. L. and Reinig, B. A., "Personalization management systems," presented at the 37th Annual Hawaii International Conference on System Sciences (HICSS'04), Hawaii, 2004.
[2] Karypis, G., "Evaluation of item-based top-N recommendation algorithms," presented at the ACM 10th International Conference on Information and Knowledge Management, Atlanta, Georgia, 2001.
[3] Gordon, T. F., "Introduction to E-government," in European Research Consortium for Information and Mathematics, vol. 48, 2002, pp. 12-13.
[4] Accenture, "eGovernment leadership: high performance, maximum value,http://www.accenture.com/xdoc/en/industries/government/gove_egov_value.pdf. Fifth Annual Accenture eGovernment Study, 2004.
[5] Eirinaki, M. and Vazirgiannis, M., "Web mining for web personalization," ACM Transactions on Internet Technology, vol. 3, pp. 1-27, 2003.
[6] Mobasher, B., Dai, H., Luo, T. and Nakagawa, M., "Discovery and evaluation of aggregate usage profiles for web personalization," Data Mining and Knowledge Discovery, vol. 6, pp. 61-82, 2002.
[7] Mulvenna, M. D., Anand, S. S. and Buchner, A. G., "Personalization on the net using web mining," Communications of the ACM, vol. 43, pp. 122-125, 2000.
[8] Sarwar, B., Karypis, G., Konstan, J. and Riedl, J., "Item-based collaborative filtering recommendation algorithm," presented at the 10th International World Wide Web Conference, Hong Kong, China, 2001
[9] Berry, M. J. A. and Lindoff, G., Data mining techniques: for marketing, sales, and customer support. New York: John Wiley & Sons, Inc,1997
