You are on page 1of 16

Paper Reading April 7

th

2015

By Oscar Li Jen Hsu


At National Tsin Hua University, Taiwan
Adomavicius, G.; Tuzhilin, A., "Toward

the next generation


of recommender systems: a survey of the stateof-the-art and possible extensions," Knowledge and
Data Engineering, IEEE Transactions on , vol.17, no.6, pp.734,749, June
2005 doi: 10.1109/TKDE.2005.99

Categories of Recommender

3 Categories

Content-based
Collaborative
Hybrid

similar to user preferred in the past.


similar to other similar user preferred in the past.

Commercial V.S. Academic

Other Categories

Rating-based

Preference-based filtering

Predict the ratings of unseen items

Categories of Algorithms in Collaborative

Memory-based

According to other similar user rating in the past.


TF-IDF

Model-based

Use collection of ratings to learn a model and to


predict ratings.
K-means cluster, Gibbs sampling, Bayesian model,
Linear regression, Maximum Entropy(ME), Markov
Decision Processes(MDP), Probabilistic Latent
Semantic analysis(PLSA)

Sparsity
Few people rating
Unusual tastes

Singular Value Decomposition(SVD)

To reduce dimensionality of sparse ratings matrices.

4 Categories of Hybrid

To combine the final results of collaborative and


content-based.

Linear combination, voting scheme.

To incorporate content-based characteristics into


collaborative.

To recommend an unpopular item by other similar


user profile who have rating info of the item.

4 Categories of Hybrid

To incorporate collaborative characteristics into


content-based.

To use latent semantic indexing(LSI) to create a


collaborative view from a group of content based
profiles.

A general unifying model that include both


content-based and collaborative characteristics.

Markov Chain Monte Carlo(MCMC)


Knowledge-based system

REINFORCED
RECOMMENDER

Comprehensive Understanding
of Users and Items
Not only ratings information but also user and
item profiles.
User demographics [69,77]
Data mining [1,34]
Sequences [63]
Signatures [26]
N nearest-neighbors
Keywords [9,76]

Extensions for Model-Based


Recommendation Techniques

Reinforced Model-Based

Radial basis functions [16,30,92]

Multidimensionality of Recommendations
To form vectors from multi-factors which like
location of item, time, company.
Hierarchical Bayesian method [5]

Multcriteria Ratings
Ratings include food, dcor, service for
restaurants.
Pareto optimal

Nonintrusiveness

Not directly to ask users to fill rating feedback,


but record user data from background, the data
like spending time.

Flexibility
The recommendation rule can be change at
running time.
A SQL command can have different WHERE
condition or can have totally different SQL
command at running time.
OLAP-based approach [2,3]

A aggregated(group of items) recommendations but


to recommend a single item.

Effectiveness of Recommendations
To evaluate recommenders.
Information Retrieval measures
F-measure
Receiver Operating Characteristic(ROC)
limited by Sample items but random items

Idea of our GPS trajectories

To recommend merchandise according to users


frequent location.

You might also like