Intelligent Search Engine and
Recommender Systems based on
Knowledge Graph
阳德青
复旦大学知识工场实验室
yangdeqing@fudan.edu.cn
2017-07-13
Background
• Knowledge Graph exhibits its
excellent performance through the
intelligent applications built on it
• As typical AI systems, Search engine
and recommender system are very
popular and promising in the era of
large data
• Many previous literatures and
systems have proved KG’s merits on
such AI’s applications
KG-based Search Engine
The History of Search Engine
• 1st: category-based
• Yahoo, hao123
• 2nd: IR-based
• Keyword-based, vector
space, Boolean model
result in
• 3rd: link-based
• PageRank (Google)
• The keyword of high click
frequency are ranked higher
• The pages containing the
keywords of more weights
are ranked higher
• The pages having more
important in-links are ranked
higher
if users want to search something new or the ones of long tail?
However, how to handle it
The History of Search Engine
• When Knowledge Graph emerges
keyword/string
thing/entity
relevant entities
The History of Search Engine
• Search 4.0: user-based search engine
• Really understand the search intent of
users, especially for those rare entities
• Directly return pure answers (entities)
instead of relevant web pages
• Provide plentiful/relevant results with
high confidence/interpretability
intelligent search engine
Knowledge graph can help the engine achieve these goals
It is actually a recommendation.
Entity Suggestion with
Conceptual Explanation [1]
• More than search
iPhone 6 Plus
or
Can any search
engine return
such results?
• Problem statement
• Given a query q described by a set of entities, the engine should return some relevant
entities along with some fine-grained concepts which can well explain the results
Entity Suggestion with
Conceptual Explanation
• What are the relevant entities and fine-grained concepts?
Growing Market
Emerging economy
BRIC
What entities should be suggested?
Country
Company
Chinese company
Chinese internet giant