The problem of information overload refers to the difficulty people have finding relevant, interesting information and making informed decisions when too much information exists. It affects people in their everyday lives, from finding interesting news articles to read, to deciding what to watch at the cinema, to finding a new restaurant to dine in. While it is not practical for newspapers, cinemas or restaurants to offer a personalised experience to every user, technological solutions are not subject to the same limitations.
Recommendation Engines have been incorporated into well known e-commerce sites, such as Amazon.com, news providers, for instance Google News and a number of entertainment recommenders including music, movies, books and television. It also appears in less obvious forms as browser plug-ins, promotional emails, pop-up advertisements and even manually-compiled lists, such as TiVo’s Guru Guides. Retailers have discovered the benefits of treating each person as an individual and providing them with a tailored surfing experience; these include improved customer loyalty and retention, increased revenue, and enhanced site navigation.
The Personalisation Process
Recommender systems are a widely accepted tool that can help users search, sort, filter and classify information, thereby helping to combat information overload. They rely entirely on the data that can be collected and utilise a five step personalisation process to generate recommendations.
The dataset is a collection of items and users (who have rated at least 1 item). The active user is the user currently targeted by the system to receive recommendations, they will have rated a subset of items. The remainder of the item set, not rated by the active user, is available for recommendation.
The personalisation process consists of five fundamental steps; data collection, pre-processing, user profile formation, pattern discovery and evaluation. It begins with data being collected, analysed and transformed into representative user behavioural profiles. These are matched against items and / or other users to produce recommendations for active users. The user profile formation and pattern discovery (steps 3 and 4) are collectively known as the recommender algorithm and result in the generation of personalised recommendations. In the final step the recommendations are evaluated to measure the success of the process. The diagram opposite shows a high level overview of the personalisation process along with the necessary inputs and desirable outputs.
Would you like to know more about the steps in the personalisation process?
How are the recommendations generated?
Recommendations can be thought of as any suggestion based on previous or current user actions. They are produced in a variety of ways; similarities between items or users (collaborative), likenesses of items to those previously purchased (content-based), classifying based on personal attributes (demographic), calculating utility functions (utility-based), using functional knowledge of how an item meets a user’s need (knowledge-based) or from a combination of these (hybrid).
Books on my shelf
Read more about recommendation techniques
- What is Collaborative Filtering?
Collaborative Filtering is an information filtering technique which is used in Recommendation Systems to find patterns within data in order to generate personalised recommendation. This Hub provides an overview of Collaborative Filtering including a
- What is Content Filtering?
Content-based filtering is a technique that originated in the information retrieval community, with email filtering and search engines being the most common applications.