|
Future Enterprise- Recommender Systems
|
| Written by David Hunter Tow |
Recommender systems are helping online consumers make purchasing decisions based on user information that is most relevant to them. Users offer feedback on purchased items and the recommender system then uses that information to predict future preferences, customised within a personal context.
Most online shopping sites and many other marketing applications now use recommender systems based on personalisation techniques. Among the most popular examples are Amazon for books and DVDs etc and Netflix which recommends movies.
However the vast majority of recommender systems still use a single criteria such as a numerical rating to represent an item’s utility. Single criteria rating systems have proved successful in some applications, but many industries have begun employing multi-criteria averaging methods- for example to establish restaurant, video and consumer electronics ratings. However these systems are not personalised- the rating being the same for all users.
Collaborative filtering recommender systems are an electronic extension of every-day social recommendations based on consumer behaviour: people share opinions and decide whether to act or not on the basis of what they hear.
Collective filtering allows scaling of interactions to groups of thousands or even millions of internet consumers. Most actual recommendation models work on profiles that have been aggregated over time by simply collecting user preference indicators, but in many domains such preferences are not static but evolve over time. Taking this dynamic reality into account in recommender model building, will introduce a more sophisticated level of performance assessment.
However other systems based on peer group recommendations as in social networking, have an advantage over collaborative systems. The recommender has a known identity that the receiver has come to trust. Anonymous users of an online system on the other hand are generally seen as less trustworthy and can vary their profiles and identities in many dofferent ways.
Recent research also highlights the lack of robustness that collaborative recommender systems exhibit against manipulation or attack, which can lead to biased advice for targeted items. There has also been recent research aimed at quantifying collaborative vulnerabilities and attempts to devise more secure defences. For example detection methods which focus on anomalous trends.
But consumer cynicism about advertising and the media is fuelling a growing reliance on word-of-mouth or person-to-person recommendations, which are the only communications medium in which trust is growing. Social networks are therefore beginning to play a vital role in marketing and highlighting the role of such support systems. Advertisers are therefore beginning to include word of mouth campaigns in their marketing mix.
Future Trends
Mechanisms that more effectively integrate the knowledge of a community of interest, combined with participative architectures that facilitate the development of integrated knowledge bases and user applications such as web portals, are now driving improvements in the quality of recommender knowledge bases.
Research in such multi-criteria problems is already extensive in both the operations research and decision science fields. Most engineering problems are essentially multi-criteria optimisation problems. Decision-making is also a multi-criteria problem that considers multiple points of view or criteria conflicting and competing with each other. However these multi-criteria problems do not apply to personalisation and recommendation contexts so that this area has remained largely unexplored.
Another future research area is the integration of theories from cognitive psychology and decision theory into recommender applications, providing new insights into the major factors influencing consumer behaviour.
And lastly to be convincing, the algorithms on which recommendations are based must be transparent and open to challenge and clearly explainable to customers..
As marketing strategies migrate increasingly to the web, linked to the rise of social communities such as Facebook and Second Life, it will be vital for the future enterprise to understand and leverage new developments in the rejuvenated recommendations field.
Only registered users can write comments.
Please login or register. |
| [ Back ] |
