Skip to content

Recommender Systems

A recommender system, or a recommendation system, is a subclass of information filtering system that seeks to predict the "rating" or "preference" a user would give to an item.

The entity to which the recommendation is provided is referred to as the user, and the product being recommended is also referred to as an item. Therefore, recommendation analysis is often based on the previous interaction between users and items, because past interests and proclivities are often good indicators of future choices.

These relations can be learned in a data-driven manner from the ratings matrix, and the resulting model is used to make predictions for target users.The larger the number of rated items that are available for a user, the easier it is to make robust predictions about the future behavior of the user.

The problem may be formulated in two ways:

  • Prediction version of problem: This approach aims to predict the rating value for a user-item combination. It is assumed that training data is available, indicating user preferences for items. For m users and n items, this corresponds to an incomplete m x n matrix, hwere the specified (or observed) values are used for training. The missing (or unobserved) values are predicted using this training model. This problem is also referred to as the matrix completion problem.
  • Ranking version of problem: This approach aims to recommend the top-k items for a particular user, or determine the top-k users to target for a particular item. Being the first one more common. The problem is also referred to as the top-k recommendation problem.

Goals of recommender systems⚑

The common operational and technical goals of recommender systems are:

  • Relevance: Recommend items that are relevant to the user at hand. Although it's the primary operational goal, it is not sufficient in isolation.
  • Novelty: Recommend items that the user has not seen in the past.
  • Serendipity: Recommend items that are outside the user's bubble, rather than simply something they did not know about before. For example, if a new Indian restaurant opens in a neighborhood, then the recommendation of that restaurant to a user who normally eats Indian food is novel but not necessarily serendipitous. On the other hand, when the same user is recommended Ethiopian food, and it was unknown to the user that such food might appeal to her, then the recommendation is serendipitous. Serendipity has the beneficial side effect of beginning new areas of interest.
  • Increase recommendation diversity: Recommend items that aren't similar to increase the chances that the user likes at least one of these items.

Basic Models of recommender systems⚑

There are four types of basic models:

  • Collaborative filtering: Use collaborative power of the ratings provided by multiple users to make recommendations.
  • Content-based: Use the descriptive attributes of the user rated items to create a user-specific model that predicts the rating of unobserved items.
  • Knowledge-based: Use the similarities between customer requirements and item descriptions.
  • Hybrid systems: Combine the above to benefit from the mix of their strengths to perform more robustly.

Collaborative Filtering Models⚑

These models use the collaborative power of the ratings provided by multiple users to make recommendations. The main challenge is that the underlying ratings matrices are sparse. To solve it, unspecified ratings are guessed by analyzing the relations of high correlation across various users and items.

There are two common types of methods.

  • Memory-based methods: Also referred to as neighborhood-based collaborative filtering algorithms, predict ratings on the basis of their neighborhoods. These can be defined through two ways:

    • User-based collaborative filtering: Ratings provided by like-minded users of a target user A are used in order to make the recommendations for A. Thus, the goal is to determine users who are similar to the target user A, and recommend ratings for the unobserved ratings of A by computing the averages of the ratings of this peer group.
    • Item-based collaborative filtering: In order to make the rating predictions for target item B by user A, the first step is to determine a set S of items that are most similar to target item B. The ratings in the item set S, which are defined by A, are used to predict whether the user A will like item B.

    These systems are simple to implement and the resulting recommendations are often easy to explain. On the other hand, they do not work very well with sparse rating matrices.

  • Model-based methods: Machine learning and data mining methods are used in the context of predictive models. In cases where the model is parameterized, the parameters of this model are learned within the context of an optimization framework. Some examples of such methods include decision trees, rule-based models, Bayesian methods and latent factor models. Many of these methods have a high level of coverage even for sparse ratings matrices.

Types of ratings⚑

The design of recommendation algorithms is influenced by the system used for tracking ratings. There are different types of ratings:

  • interval-based: A discrete set of ordered numbers are used to quantify like or dislike.
  • ordinal: A discrete set of ordered categorical values, such as agree or strongly agree, are used to achieve the same goal.
  • binary: Only the like or dislike for the item can be specified.
  • unary: Only liking of an item can be specified.

Another categorization of rating systems is based in the way the feedback is retrieved:

  • explicit ratings: Users actively give information on their preferences.
  • implicit ratings: Users preferences are derived from their behavior. Such as visiting a link. Therefore, implicit ratings are usually unary.

Content-Based Recommender Systems⚑

In content-based recommender systems, descriptive attributes of the user rated items are used to create a user-specific model that predicts the rating of unobserved items.

These systems have the following advantages:

  • Works well for new items, when sufficient data is not available. If the user has rated items with similar attributes.
  • Can make recommendations with only the data of one user.

And the following disadvantages:

  • As the community knowledge from similar users is not leveraged, it tends to reduce the diversity of the recommended items.
  • It requires large number of rating user data to produce robust predictions without overfitting.

Knowledge-based Recommender Systems⚑

The recommendation process is performed based on the similarities between user requirements and item descriptions or the user requirements constrains. The process is facilitated with the use of knowledge bases, which contain data about rules and similarity functions to use during the retrieval process.

Knowledge-based systems can be classified by the type of user interface:

  • Constraint-based recommender systems: Users specify requirements on the item attributes or give information about their attributes. Then domain specific rules are used to select the items to recommend.
  • Case-based recommender systems: Specific cases are selected by the user as targets or anchor points. Similarity metrics are defined in a domain specific way on the item attributes to retrieve similar items to these cases.

These user interfaces can interact with the users through several ways:

  • Conversational systems: User preferences are determined iteratively in the context of a feedback loop. It's useful if the item domain is complex.
  • Search-based systems: User preferences are elicited by using a preset of questions.
  • Navigation-based recommendation: Users specify a number of attribute changes to the item being recommended. Also known as critiquing recommender systems.

The main difference between content-based systems and knowledge-based systems is that while the former learns from past user behavior, the latter does it from active user specification of their needs and interests.

These systems have the following advantages:

  • Works well for items with varied properties and/or few ratings. Such as in cold start scenarios, if it's difficult to capture the user interests with historical data or if the item is not often consumed.
  • Allows the users to explicitly specify what they want, thus giving them a greater control over the recommendation process.
  • Allows the user to iteratively change their specified requirements to reach the desired items.
  • Can make recommendations with only the data of one user.

And the following disadvantages:

  • As the community knowledge from similar users is not leveraged, it tends to reduce the diversity of the recommended items.

Domain-Specific recommender systems⚑

Demographic recommender systems⚑

In these systems the demographic information about the user is leveraged to learn classifiers that can map specific demographics to ratings.

Although they do not usually provide the best results on a standalone basis, they enhance and increase robustness if used as a component of hybrid systems.

Pitfalls to avoid⚑

Pre-substitution of missing ratings is not recommended in explicit rating matrices as it leads to a significant amount of bias in the analysis. In unary ratings it's common to substitute the missing data by 0 as even though it adds some bias, it's not as great because it's assumed that the default behavior.

Interesting resources⚑

Bookworm looks to be a promising source to build book recommender systems.

Content indexers⚑

  • Open Library: Open, editable library catalog, building towards a web page for every book ever published. Data can be retrieved through their API or bulk downloaded.

Rating Datasets⚑

Books⚑

  • Book-Crossing: 278,858 users providing 1,149,780 ratings (explicit / implicit) about 271,379 books.

Movies⚑

Music⚑

Web⚑

Miscelaneous⚑

  • Wikilens: generalized collaborative recommender system that allowed its community to define item types (e.g. beer) and categories (e.g. microbrews, pale ales, stouts), and then rate and get recommendations for items.

Past Projects⚑

  • GroupLens: Pioneer recommender system to recommend Usenet news.
  • BookLens: Books implementation of Grouplens.
  • MovieLens: Movies implementation of Grouplens.

References⚑

Books⚑

  • Recommender Systems by Chary C.Aggarwal.
  • Recommender systems, an introduction by Dietmar Jannach, Markus Zanker, Alexander Felfernig and Gerhard Friedrich.
  • Practical Recommender Systems by Kim Falk.
  • Hands On recommendation systems in Python by Rounak Banik.

Awesome recommender systems⚑