Discovery of Aggregate Usage Profiles for Web Personalization

1
Discovery of Aggregate Usage Profiles for Web
Personalization

• Bamshad Mobasher, Honghua Dai, Tao Luo, Miki
  Nakagawa, Jim Wiltshire

School of Computer Science, Telecommunications,
and Information Systems DePaul University
2
Web Personalization

• The Problem
• dynamically serve customized content (pages,
  products, etc.) to users based on their profiles,
  preferences, or expected interests
• Current Approaches
• rule-based filtering
• usually relies on static profile for users in
  part obtained through explicit registration
• collaborative filtering
• usually requires explicit ratings from users on
  similar types of objects
• content-based filtering learn/store personal
  profiles locally or on server-side
• based on content similarity of user profile to
  pages or product descriptions
• Limitations of Current Technologies
• user input may be subjective and prone to bias
• explicit (and non-binary) user ratings may not be
  available
• profiles may be static and can become outdated
  quickly
• collaborative filtering problems with
  scalability due to sparse data
• content-based filtering may miss other semantic
  relationships among objects

3
Usage-Based Web Personalization

• Basic Idea
• find aggregate user profiles by automatically
  discovering user access patterns through Web
  usage mining (offline process)
• data sources for mining include server logs,
  other click-stream data (e.g., product-oriented
  user events), and site structure
• match a users active session against the
  discovered profiles to provide dynamic content
  (online process)
• Advantages / Goals
• profiles are based on objective information (how
  users actually use the site)
• no explicit user ratings or interaction with
  users (to enter a profile, etc.)
• helps preserve user privacy, by making effective
  use of anonymous data
• usage data captures relationships missed by
  content-based approaches
• can help enhance the effectiveness of
  collaborative or content-based filtering
  techniques

4
Automatic Web PersonalizationOffline Process
Data Preparation
Usage Mining
Transaction Clustering Pageview Clustering
Usage Profiles
Data Cleaning Session Identification Pageview
Identification Transaction Identification Support
Filtering
Server Logs Other Click-Stream Data
Association-Rule Discovery
Domain Knowledge
5
Automatic Web PersonalizationOnline Process
Recommendation Engine
Input from the batch process
Recommendations
Active Session
6
Data Preparation Tasks

• Preprocess and filter logs and other usage data
• remove redundant references and create pageviews
• domain knowledge to assign types to pageviews
• handle references to scripts creating dynamic
  pages
• map logs against site topology
• Identify user sessions and transactions
• heuristics based on IP, referrer, agent fields,
  and session time-outs used to identify unique
  user sessions (may need to infer missing
  references)
• intra-session transactions can be obtained based
  on a model of user behavior (involves classifying
  references as content or navigational for
  each user)
• weights are assigned to each pageview based on
  static pageview types as well as some measure of
  user interest (e.g., duration of pageview)
• Support filtering - remove very low/high support
  pageviews

7
Aggregate Usage Profiles

• Characteristics of Aggregate Profiles
• the goal is to effectively capture common usage
  patterns from potentially anonymous click-stream
  data
• profiles are represented as weighted collections
  of pageviews
• weights represent the significance of pageviews
  within each profile
• profiles are overlapping in order to capture
  common interests among different groups/types of
  users
• multiple profiles may contribute to the
  recommendation set for a given user
• Example Profiles from the ACR (Assoc. for
  Consumer Research) Site

1.00 Call for Papers 0.67 ACR News Special
Topics 0.67 CFP Journal of Psychology and
Marketing I 0.67 CFP Journal of Psychology and
Marketing II 0.67 CFP Journal of Consumer
Psychology II 0.67 CFP Journal of Consumer
Psychology I
1.00 CFP Winter 2000 SCP Conference 1.00 Call
for Papers 0.36 CFP ACR 1999 Asia-Pacific
Conference 0.30 ACR 1999 Annual
Conference 0.25 ACR News Updates 0.24 Conference
Update
8
Methodologies for the Discovery of Aggregate
Profiles

• Discovery of Profiles Based on Transaction
  Clusters
• cluster user transactions - features are
  significant pageviews identified in the
  preprocessing stage
• derive usage profiles (set of pageview-weight
  pairs) based on characteristics of each
  transaction cluster
• Cluster Pageviews
• directly compute overlapping clusters of
  pageviews based on co-occurrence patterns across
  transactions
• features are user transactions, so dimensionality
  poses a problem for traditional clustering
  algorithms
• we use Association-Rule Hypergraph Partitioning
  with an overlap factor

9
Profile Aggregation Based on Clustering
Transactions (PACT)

• Input
• set of relevant pageviews in preprocessed log
• set of user transactions
• each transaction is a pageview vector
• Transaction Clusters
• each cluster contains a set of transaction
  vectors
• for each cluster compute centroid as cluster
  representative
• Aggregate Usage Profiles
• a set of pageview-weight pairs for transaction
  cluster C, select each pageview pi such that
  (in the cluster centroid) is greater than a
  pre-specified threshold

10
Hypergraph-Based Clustering

• Construct a hypergraph from sets of related items
• Each hyperedge represents a frequent itemset
• Weight of each hyperedge can be based on the
  characteristics of frequent itemsets or
  association rules

• Recursively partition hypergraph so that each
  partition contains only highly connected data
  items
• Given a hypergraph G(V,E) we find a k-way
  partitioning such that the weight of the
  hyperedges that are cut is minimized
• The fitness of partitions measured in terms of
  the ratio of weights of cut edges to the weights
  of uncut edges within the partitions
• The connectivity measures the percentage of edges
  within the partition with which the vertex is
  associated -- used for filtering partitions
• Vertices from partial edges can be added back to
  clusters based on a user-specified overlap factor

11
Profiles Based on Hypergraph Clusters of Pageviews

• Input
• input for clustering is the set of large itemsets
  from association rule module
• each itemset is a hyperedge (weights are a
  function of the interest of the itemset)

• Aggregate Profiles (Pageview Clusters)
• hMETIS used as the underlying hypergraph
  partitioning algorithm
• clustering program directly outputs a set of
  overlapping pageview clusters
• the weight associated with pageview p in a
  cluster C is based on the connectivity value of p
  in hypergraph partition

12
Recommendations Based on Usage Profiles

• Match current users activity against the
  discovered usage profiles
• a sliding window over the active session to
  capture the current users short-term history
  depth
• usage profiles and the active session are treated
  as vectors
• matching score is computed based on the
  similarity between vectors (e.g, normalized
  cosine similarity)
• Recommendations
• each pageview is assigned a recommendation score
  based on
• matching score to aggregate profiles
• information value of the pageview based on
  domain knowledge (e.g., link distance of the
  candidate recommendation to the active session)
• recommendations are contributed by multiple
  matching aggregate profiles

13
Experimental Set-up

• The Data Sets
• Log data from the Association for Consumer
  Research Web site
• 18342 transactions, 62 pageview URLs (after
  filtering)
• Data set divided into training and evaluation
  sets
• Evaluation Methodology
• Portion of each transaction (based on a specified
  window size) in evaluation set was used to
  generate a recommendation set (based on a given
  recommendation threshold)
• For each transaction, the overall coverage of the
  recommendation set was divided by the number of
  recommendations to produce an accuracy measure
• The overall score was computed (for each
  threshold) by taking the average scores over all
  transactions in the evaluation set

14
Average Visit Percentage
AVP measures the likelihood that a user who
visits any page in a Given profile, also visits
other pages in that profile
15
Evaluation Measuring Recommendation Accuracy
Recommendation accuracy results, using a active
session window of size 3.
16
Evaluation Impact of Filtering
Comparison of PACT and Hypergraph (using window
size 2) for filtered and unfiltered data sets.
Filtering involved the removal of top-level
navigational pages from the data set, leaving
only deeper content-oriented pages.
17
Conclusions

• Usage-Based Web Personalization
• results suggest that effective personalization
  can be achieved even with anonymous and
  short-term click-stream data
• possibly useful in the early stages of
  personalization when more detailed profiles are
  not available for individual users
• could be used effectively in conjunction with
  other methods based on content-based or
  collaborative filtering
• Which Method is Best?
• PACT may be most appropriate when the goal is to
  provide a more general personalization solution
  involving a variety of objects across the whole
  site
• Hypergraph may be most appropriate when the goal
  is to provide a highly focused set of
  recommendations for specific portions of the site
• In practice, usage-based methods need to be
  combined with other techniques to provide an
  integrated solution