Modern Web Search with Consideration of User Experience
During the recent courses, we’ve thoroughly and detailedly discussed about abstract networks — the Internet and the web. In my view, the topic about ranking the results we get from querying the web is extremely useful to modern people living in such an intensively information-based society. In the class, the professor introduced the PageRank algorithm, that is, we originally give each web-page equal page rank, and then gradually update the rank a large amount of times to be the sum of the shares it receives from other pages. However, in the era of rapidly developed information system, only using number of times of being referred to weight the page rank of each webpage cannot precisely predict the preference or expectation of customers. With the popularization of machine learning, combining the previous usage data of each customer to generate an individual and personalized page rank for each customer.
https://link.springer.com/content/pdf/10.1007%2Fs11257-011-9112-x.pdf
In the paper Recommender systems: from algorithms to user experience published by Joseph A. Konstan and John Riedl, authors specifically discussed about the evolutionary advances in collaborative filtering recommender systems, which previously focusing on algorithms’ effectiveness and accuracy then gradually adding more consideration of user experience. In the first part of the paper, we can have a look at the algorithms of reference by combining users’ data (as suggested in the book chapter 14.4 about APPLYING LINK ANALYSIS IN MODERN WEB SEARCH). In pure algorithms’ development, using user-user collaborative filtering( massively storing users’ purchase histories and preferences to predict the taste of other users) is comparatively unsuccessful and memory-consuming than using item-item collaborative filtering (storing the correlation between two items that when people purchase one, the likelihood of purchasing the other). That’s reasonable because of the diversity and broadness of customers’ preferences.
However, the later part of the article counterintuitively introduces the notion that storing the users’ experience, that is the lifecycle of user-recommender is more effective and worthwhile. By carefully studying the development of users’ behaviors from a fresh-user to an experience one, including how recommender systems can adapt to different needs of new users vs. experienced users, and how they can balance short-term with longer-term value, researchers add one more consideration to our original Page Rank algorithms. With the more advanced model, the Internet can better and more swiftly “understand” each customers.
Nevertheless, there’s also drawback of the research, the more complexed model scientists have developed, the more dangerous environment our privacies will be exposed in. While the prediction algorithm can exactly refer the item we’re aspiring for, we customers are, at the same time, being controlled and invisibly manipulated by the seller. Such kind of mindless disclose of personal data is putting each one’s information in a transparent state, while we don’t know what and how others might do with our private information.
I think this article is an extension of the page ranking algorithm we’ve learnt in the class. By introducing the development of page ranking algorithms which takes more aspects into consideration, the authors enable readers to not only learn the evolution but also the potential drawback of more advanced algorithms.