3.4_Recommender.md

3.4 Introduction to Recommender Systems

Introduction

• Given a user model (ratings, preferences, ...) and items, find relevance score
• Typically used for ranking

Paradigms of Recommender Systems

• Personalized recommendations: user profile & contextual parameters
• Collaborative: community data
• Content-based: product features
• Knowledge-based: knowledge models
• Hybrid: combinations of various inputs and/or composition of different mechanism

Collaborative Filtering

Pure CF Approaches

• Input: matrix of given user-item ratings
• Output types: degree to what the current user will like or dislike a certain item; a top-N list of recommended items

User-based nearest-neighbor CF

Basic algorithm

• given an "active user" $u$ and an item $i$ not yet seen by $u$
  • find a set of users (peers/nearest neighbors) who liked the same items as $u$ in the past and who have rated item $i$
  • combine their ratings to predict if $u$ will like item $i$
• repeat this for all items that $u$ has not seen
• Recommend the best-rated items

Basic assumptions

• if users had similar tastes in the past, they will have similar tastes in the future
• user preferences remain stable and consistent over time

Evaluation

• Business questions lead to empirical evaluation during development and in deployment

Offline Evaluation

• Data: collected in your problem; benchmark datasets
• Remove useless data:
  • training set: randomly selected share of known users
  • testing set: recommendations based on observed items compared to hidden items

Metrics

• Precision: exactness; fraction of relevant items retrieved out of all items retrieved

• Recall: completeness; fraction of relevant items retrieved out of all relevant items

  • When tuning system to increase precision, recall decreases

• F1 Metric: combine Precision and Recall into a single value

• Rank Metrics: extend recall and precision to take the positions of correct items in a ranked list into account

• Rank Score: extend recall metric to take the positions of correct items in a ranked list into account

Normalized Discounted Cumulative Gain

• Discounted cumulative gain (DCG): logarithmic reduction factor

• Idealized discounted cumulative gain (IDCG): assumption that items are ordered by decreasing relevance

• Normalize discounted cumulative gain (nDCG): normalized to the interval [0...1]

• Average Precision: ranked precision metric that places emphasis on highly ranked correct predictions (hits)

Metrics for Rating Prediction

• Ground truth = ratings
• Mean Absolute Error (MAE): computes the deviation between predicted ratings and actual ratings
• Root Mean Square Error (RMSE): similar to MAE, but places more emphasis on larger deviation

Online Evaluation

Characteristics of methods

• Subject
• Research Method
• Setting

Evaluation Settings

Lab Studies

• Expressly created for the purpose of the study
• Extraneous variables can be controlled more easily by selecting study participants who should behave as they would in a real-world environment but doubts may exist about participants motivated by money, prizes or social pressure

Field Studies

• Conducted in a preexisting real world environment
• Users are intrinsically motivated to use a system

Experimental Design

• Hypotheses on personalized vs. non-personalized recommendation techniques and their potential to do something

Non-experimental Research

• Quasi-experiments: lack random assignments of units to different treatments
• Non-experimental/observational research: surveys/questionnaires, longitudinal research, case studies, focus group

Data

Types of Ratings

Explicit

• typical choices
• possibly multidimensional
• main challenge: users not always willing to rate many items

Implicit

• user action interpreted as rating
• easy to collect transparently without additional effort
• main challenge: action doesn't necessarily have the same meaning as a rating

Sparsity

• Data is sparse
• Natural datasets include historical interaction records of real users
• Sparsity can be measured $Sparsity = 1 - |R|/|I|.|U|$, R = ratings, I = Items, U = Users

Problems

• How many items in common are 2 users expected to have?
• Cold start: How to recommend new items? What to recommend to new users
  • Ask/force users to rate a set of items
  • In the beginning use method not based on rating
  • Default voting

More Algorithms

Memory-based approaches

• User-based CF
  • rating matrix is directly used to find neighbors/make predictions
  • does not scale for most real-world scenarios

Model-based approaches

• Based on an offline pre-processing or "model-learning" phase
• At run-time: only the learned model is used to make predictions
• Models are updated/re-trained periodically
• Matrix factorization techniques, statistics
• Association rule mining
• Probabilistic models (clustering models, Bayesian networks, probabilistic Latent Semantic Analysis)
• Various other machine learning approaches

Content-based recommendation

• Content: combination of attributes and (semi-)free text
  • Recommendation approach: related to NLP and document classification

Knowledge-based recommendations

• Users want to define their requirements explicitly
• Time span plays an important role
• Items with low number of available ratings
• Constraint-based: based on explicitly defined set of recommendation rules; fulfill recommendation rules
• Case-based: based on different types of similarity measures; retrieve items that are similar to specified requirements
• Both approaches are similar in their conversational recommendation process
  • users specify the requirements
  • systems try to identify solutions
  • if no solution can be found, users change requirements

Interaction: critiquing

• User may not know exactly what they are seeking but can specify why their current item is not satisfactory

Hybrid Recommender Systems

• Monolithic exploitation of different features
• Parallel
• Pipeline

Challenges

Explanation

2 parties involved:

• organization interested in convincing user
• user concerned about making the right choice(s)

Attacks

• (monetary) value of being in recommendation lists
• Attacks aim to:
  • push some items
  • sabotage other items
  • simply sabotage the system
  • manipulation the "internet opinion"

Research Questions in Ubiquitous Domains

• Goals: serendipitous recommendations vs. proximity
• role of contextual parameter
• modality of interaction for users "on the go"

Application domains

• M-Commerce
• Tourism and visitor guide
• Cultural heritage and museum guides
• Home computing and entertainment

< Go back