1.1 Background and Motivation
Online social networking systems and peer-produced services have gained much attention as a social medium of viral marketing, which exploits existing social networks by inspiring bloggers to share their own posts or personal information with the other bloggers. The weblogs indeed provide a more open channel of communication for people in the blogosphere to read, commentate, cite, socialize and even reach out beyond their social networks, make new connections, and form communities [10]. A blog social network has emerged as a powerful and potentially services-valued form of computer-mediated communication (CMC).
More and more interactions take place in the blogosphere, combining the benefits of the accessibility of the web, the ease-of-use of interface and the incentive of blogging (i.e. share, recommend, comment…etc.). Blog becomes a viral marketing site based on peer-production and it is promoted yet induced by online person to person interactions. Moreover, there exists a large number of information in the blogosphere, including text-based blog entries (articles) and profile, pictures or figures, and multimedia resources. This becomes problematic for users.
How do they deal with information overload problems and how do they effectively retrieve information they consider important? This gives us an incentive to develop a blog recommender approach and design an information filtering mechanism.
A recommender system of weblog differs from others in several ways. First, a recommendation target varies dramatically from product, movie, music, news, webpage, travel and tourism to all kinds of service, online auction seller, or even virtual community [12].
It is important for us to find the characteristics of recommendation targets because inappropriate use of recommendation may have a totally opposite effect by resulting unfavorable attitudes towards the recommendation target. Second, the blog recommender is
also a provider. Unlike other contexts, blogs or bloggers in the entire blog network are highly dynamic and the recommendation environment changes fast. The blog recommendation mechanism must be more flexible and adaptable than the others. Third, it is more human-oriented. In other words, blog content itself is highly subjective and textual-sensitive for recommenders.
Blog search engine and blog recommender system serve similar function but differ to some extent. What is the difference between blog search engine and blog recommender system? This question emerges as the blog filtering approach such as search engine can also alleviate the mentioned problem. There are three folds of differences between them. First, information needs: real-time versus long-run. Some weblog aggregators, such as Technorati, provides tag-based search engine platform; moreover, Blogpulse and Daypop supply common keyword-based search engines just like Google and Yahoo but are applied in weblog domain.
It allows users to find potential interesting postings, which many bloggers are talking and concerning about recently, with ease. In contrast to search engine technology, the proposed blog recommendation mechanism is long-run oriented. In other words, the former focuses on popularization however the latter is more personalized. Second, pull versus push information:
the former is a paradigm of technology for pulling information. A search result is obtained after the query is submitted. As for the latter, either pull or push technology could be employed to induce the recommendation results. Third, diversity of recommendation process:
the former only considers the content and term comparison. As for the latter, it considers multidimensional approaches and factors to implement the recommendation mechanism. In this study, the proposed mechanism takes all these three factors into consideration.
Moreover, recommender system is a useful alternative to search algorithms, since they help users discover items they might not have found by themselves. Interestingly enough, recommender systems are often implemented using search engines indexing data. That is, some recommender systems are proposed based on the results of search engine. Since search
engine could not provide personalized results according to user’s preference, a recommender mechanism will do by integrating more methodologies to make a personalized resource-provided mechanism.
1.2 Research Problems
In blog recommendation context, it is important that how we introduce interesting, personalized and socially related weblogs of this peer-produced information to bloggers through recommendation mechanism. The objective of blog recommendation mechanism in this study is bloggers or blog posts (articles). The problem is that what kinds of blog posts do we recommend? Is it most popular? Is it most trustworthy? Or recommend most similar in links or in blog content? These approaches and related researches inspire us to combine them to propose a synthetical recommendation mechanism in this study. We believe that trust model, social relation and semantic similarity play an important role in trust recommender system, social networking analysis, and information retrieval/textual comparison respectively. They are three crucial factors to help prepare the ground for the development of personalized and trustworthy recommendation mechanism.
1.3 Research Objectives
In this research, we propose a personalized, trustworthy, and adaptive blog recommendation mechanism which integrates the trust model, social relation and, semantic analysis to construct a comprehensive model in recommending bloggers and blog posts. With this mechanism, a blogger could have better opportunities to locate more interested, trustworthy, and related blogging information with greater satisfactions than other existing recommendation approaches. More specifically speaking, we want to provide bloggers with
more precise and more desirable blogging information with less efforts and greater satisfactions.
The main objective of this research is to apply the proposed recommendation mechanism to the real-world blog platform and investigate the recommendation performances with an empirical validation. We take a famous BSP (Blog Service Provider), Wretch [26], as our target of experiments, and compare the recommendation performances with existing approaches, to examine if our proposed mechanism outperforms the existing ones.
1.4 Thesis Outline
The rest of paper is organized as follows. Section 2 presents related works. Section 3 designs a system framework of neural network based recommendation mechanism. Section 4 describes the methodologies of trust model, social relation and semantic analysis. Section 5 proposes an experimental study to discover some characteristics of blog network and demonstrates the effectiveness of the proposed recommendation mechanism. Section 6 concludes the paper, discusses the potential problems and some limitations, and describes the future works.