User Loyalty Articles

Due to the low amount of user participation and, as a result, the relatively small amount of content, it was decided to use content personalisation in this research. Collaborative filtering would have been the wrong choice as it finds a profile of another user that is most similar to the users, but without many members it is unlikely to find peers that are similar enough to provide valuable results (Adomavicius et al. 2003). Furthermore, the news section has only a small amount of new information every day, which reduces the options for success of collaborative filtering even more. Another reason for using content personalisation were the findings of Shepherd et al. (2001) that users do not enjoy too detailed information filtering, therefore the news is only filtered according to news categories.

The personalisation features were implemented in the start page, the news page and the page on which the user rates ideas. Every member was identified by his or her login and password. Because A/B testing was used, every second member that registered surfed the site with personalisation features enabled (B+user) while the rest surfed the site without them (B-user).

The implicit data for the personalisation system derived from predefined PHP variables (PHP Group, 2003), which save user data that is sent to the server, and the tracking of users’ browsing behaviour. The system is based on a user profile in which all data is saved and every site that has personalised content on it is controlled by the user profile. The user profile consists of data for identification (Identification Profile), system data such as ip address or id (System Profile), self-categorised data such as age or gender (Socio-economic Profile) and data about the actions of the user on the site (Interaction Profile)(Schubert et al. 2000)(refer to Appendix J). Every time a user loads a page his user profile is updated in the appropriate category of interest. The start page is excluded because it shows all categories of content in five modules and therefore does not represent a specific interest. Over a period of time a profile of the users’ interest in the site is created. According to this profile the ratio of each content category of the start page is adjusted. If one category was not visited yet a link is shown with a call to visit this category while B-users did not see this feature. [P4] This is done to create awareness of existing features on the site and thereby increase the user experience (McMullin 2004). Furthermore, the start page shows special features and articles according to the operating system of the user and every user is greeted according to the time of the day and the time he has not[P5] visited the site. (pic1)

On the news page the news posts are sorted by starting with news from the category the user preferred most in the past. B-users see the news items according to the order in which they were published. The news is categorised according to the five news groups that were determined in the user-perception test. These five groups have subgroups with more specific topics (refer to Appendix I). When users click on the link to the news source their profile is updated in the corresponding category. To prevent the profile being updated several times when the news has more than one link, the user profile logs every news item a user has already read and compares this data with the incoming data from every update. Only the news that was posted since the user’s last visit or in the past six hours is sorted according to the user’s profile. This is done because changing all news in real time could confuse a user and harm the user experience. To sort the news the ratio of the viewed news of a user to the total amount of news in this category was calculated. This had to be done because not every category had the same amount of news and therefore the pure data in the user profile did not have to represent the user’s interest.

The user rank was also derived from their activity on the site and favourite news category. The activity is the sum of all values of the content categories in the user profile. (pic user profile structure)

The user rank feature is the only personalisation feature that uses explicit data to a minimal degree. The gender that users assign to themselves is used to give the title the right grammatical form.

Ideas are also categorised (refer to Appendix K). On the idea-rating page a B+ user will see an idea from a category in which he rated an idea before if he did not already rate all ideas of this category. These 100%, which represent all ratings he made, are divided according to the sum of all ratings in each category. Therefore if a user rates ideas from a particular [EB6] category highly, it is more likely that they will get more ideas from this category to rate next. In this personalisation feature no ratio is necessary because the ratings control the level of interest.

The personalisation features have multiple goals. In the short term they should provide users with a better user experience on the website. The long-term goal is to give the website a personal touch and thereby establish an emotional bond between the user and the website, which results in increased loyalty.