Cheat sheet AIS
User modelling
User Modeling is the process of creating and updating a user model by deriving user
characteristics from user data —which is data that is explicitly provided by the user, or data
that stems from indirect events and observations
User modelling process
1. Acquisition of user data (p4 voor + nadelen expl/ impl estimates)
Asking the user (direct input)
More complex way is by observing the user’s actions
Using stereotypes: any kind of group or class we can identify of users. Adapt to
groups instead of individual persons. E.g. teenage user; to cover cold start, but
oversimplification.
2. Inference of knowledge from the
data: process of interpreting
events and observations on a U,
making use of conditions, rules or
other forms of reasoning, and the
storage of the inferred knowledge
in the user model.
Detecting patterns in user
behavior
Matching user behavior
with the behavior of other
users
Classifying users or content based on user behavior
3. Representation of the user model (Overlay models p5)
4. Updating a user model
Not always done, but is recommended as user changes over time
Baye’s Theorem (p7)
Context (p16)
Degree of context-awareness: Based on user interaction—from high to low:
1. User adaptation (user is active)
2. Passive context-awareness ; System constantly monitors the environment and offers
appropriate options to users
3. Active context-awareness;
System continuously and
autonomously monitors situation
and acts autonomously
Content-based filtering (p22, p38 (dis)adv)
recommends items to users based on the
characteristics or content of the items
and the user's historical preferences.
It relies on the idea that if a user has shown
interest in certain attributes or features
of items in the past, they are likely to be
interested in items with similar attributes
in the future. p23
1
, To implement content-based filtering, each item is described using a set of attributes or features.
These attributes could be keywords, tags, genres, or other relevant characteristics, depending on
the type of items (e.g., movies, books, products).
User profiles are created based on their past interactions and preferences. These profiles capture
the user's preferences for certain item attributes.
When making recommendations, the system selects items that match the user's profile by
identifying items with attributes similar to those the user has shown interest in.
Content-based filtering is especially useful when there is limited user interaction data or when
recommendations need to be explainable because it is based on item characteristics and
user preferences.
Item-based CF (p22)
aka item-item collaborative filtering, makes recommendations by identifying similarities between
items rather than focusing on user profiles.
It relies on the principle that if a user has liked or interacted with a particular item, they are likely to
be interested in items that are similar to the ones they have previously liked.
The system builds an item-item similarity matrix, which quantifies the similarity between pairs of
items based on user interactions and ratings.
When a user expresses interest in an item (e.g., by rating it positively or adding it to their list), the
system looks for similar items in the similarity matrix and recommends those similar items to the
user.
Item-based collaborative filtering is particularly effective when there is plenty user interaction
data, and it doesn't require detailed user profiles.
Problems (cold start, latency, sparseness of matrix, scalability, diverseness, privacy and trust,
changing user interests) (p24)
Collaborative filtering (p26, (dis)adv p29, verschil CF en CB p29)
System recommends items which were preferred by similar users in the past. Ratings
express preferences of the active user and also other users’ collaborative approach
Works on user-item matrix; Memory- or model-based
Assumption: Similar taste in the past implies similar taste in future
Memory vs Model-based CF (p30)
Memory based: relies on the direct memory of user-item interactions, (user-item matrix)
Model-based CF: using the user-item matrix to build a predictive model. (underlying &latent factors)
(p31)
Collaborative vs individual (content) (p32)
Term-Frequency - Inverse Document Frequency (p34)
TF-IDF: a technique used in information retrieval and text mining to represent and evaluate the
importance of terms (keywords) within a collection of documents
Knowledge-based Recommendations (p39)
Constraint-based recommender systems use explicitly defined rules to retrieve items that fulfill
user requirements, while case-based systems use similarity measures to find items similar to
user preferences.
Hybrid approaches (p41)
Weighted: Score of different recommendation components are combined (numerically)
Switching: System chooses among recommendation components and applies the selected one
Mixed: Recommendations from different recommenders are presented together
Monolithic (p43): exploiting different features: a single recommendation algorithm is employed,
but it leverages different features or types of information to make recommendations; a single,
comprehensive system that takes advantage of multiple sources of information.
2
User modelling
User Modeling is the process of creating and updating a user model by deriving user
characteristics from user data —which is data that is explicitly provided by the user, or data
that stems from indirect events and observations
User modelling process
1. Acquisition of user data (p4 voor + nadelen expl/ impl estimates)
Asking the user (direct input)
More complex way is by observing the user’s actions
Using stereotypes: any kind of group or class we can identify of users. Adapt to
groups instead of individual persons. E.g. teenage user; to cover cold start, but
oversimplification.
2. Inference of knowledge from the
data: process of interpreting
events and observations on a U,
making use of conditions, rules or
other forms of reasoning, and the
storage of the inferred knowledge
in the user model.
Detecting patterns in user
behavior
Matching user behavior
with the behavior of other
users
Classifying users or content based on user behavior
3. Representation of the user model (Overlay models p5)
4. Updating a user model
Not always done, but is recommended as user changes over time
Baye’s Theorem (p7)
Context (p16)
Degree of context-awareness: Based on user interaction—from high to low:
1. User adaptation (user is active)
2. Passive context-awareness ; System constantly monitors the environment and offers
appropriate options to users
3. Active context-awareness;
System continuously and
autonomously monitors situation
and acts autonomously
Content-based filtering (p22, p38 (dis)adv)
recommends items to users based on the
characteristics or content of the items
and the user's historical preferences.
It relies on the idea that if a user has shown
interest in certain attributes or features
of items in the past, they are likely to be
interested in items with similar attributes
in the future. p23
1
, To implement content-based filtering, each item is described using a set of attributes or features.
These attributes could be keywords, tags, genres, or other relevant characteristics, depending on
the type of items (e.g., movies, books, products).
User profiles are created based on their past interactions and preferences. These profiles capture
the user's preferences for certain item attributes.
When making recommendations, the system selects items that match the user's profile by
identifying items with attributes similar to those the user has shown interest in.
Content-based filtering is especially useful when there is limited user interaction data or when
recommendations need to be explainable because it is based on item characteristics and
user preferences.
Item-based CF (p22)
aka item-item collaborative filtering, makes recommendations by identifying similarities between
items rather than focusing on user profiles.
It relies on the principle that if a user has liked or interacted with a particular item, they are likely to
be interested in items that are similar to the ones they have previously liked.
The system builds an item-item similarity matrix, which quantifies the similarity between pairs of
items based on user interactions and ratings.
When a user expresses interest in an item (e.g., by rating it positively or adding it to their list), the
system looks for similar items in the similarity matrix and recommends those similar items to the
user.
Item-based collaborative filtering is particularly effective when there is plenty user interaction
data, and it doesn't require detailed user profiles.
Problems (cold start, latency, sparseness of matrix, scalability, diverseness, privacy and trust,
changing user interests) (p24)
Collaborative filtering (p26, (dis)adv p29, verschil CF en CB p29)
System recommends items which were preferred by similar users in the past. Ratings
express preferences of the active user and also other users’ collaborative approach
Works on user-item matrix; Memory- or model-based
Assumption: Similar taste in the past implies similar taste in future
Memory vs Model-based CF (p30)
Memory based: relies on the direct memory of user-item interactions, (user-item matrix)
Model-based CF: using the user-item matrix to build a predictive model. (underlying &latent factors)
(p31)
Collaborative vs individual (content) (p32)
Term-Frequency - Inverse Document Frequency (p34)
TF-IDF: a technique used in information retrieval and text mining to represent and evaluate the
importance of terms (keywords) within a collection of documents
Knowledge-based Recommendations (p39)
Constraint-based recommender systems use explicitly defined rules to retrieve items that fulfill
user requirements, while case-based systems use similarity measures to find items similar to
user preferences.
Hybrid approaches (p41)
Weighted: Score of different recommendation components are combined (numerically)
Switching: System chooses among recommendation components and applies the selected one
Mixed: Recommendations from different recommenders are presented together
Monolithic (p43): exploiting different features: a single recommendation algorithm is employed,
but it leverages different features or types of information to make recommendations; a single,
comprehensive system that takes advantage of multiple sources of information.
2
