07 Feedback Flashcards
what are the information needs
the underlying cause of the query that a person submit to the search engine
- can be categorised into type of information or type of task
how is query related to information need
queries is a representation off very different information needs
- but it can be a poor representation as users find it difficult to express
- encouraged to enter short queries
- ambiguous: the same query may represent different needs
what is the query formulation problem
difficult to generate well formulated queries without
- knowledge of collection
- knowledge of retrieval environments
can learn which one is relevant, first query is trial run
what is the key aspect of effective retrieval
users cant change the ranking algorithm but can change the result through interaction. IR is an iterative process
what is the ASK hypothesis
you don’t know what you don’t know:
you can’t search for it if you don’t know about it, therefore there is a need for search engines to show users a list of relevant documents
what are some examples of explicit interaction
- relevance feedback
- allow user to provide feedback, add terms from relevant document into query
- use feedback information to reformulate query and produce result
- allow more interactive process - query expansion
- query term suggestion
how does relevance feedback using vectors work
- document that are relevant to one another resemble one another in similar vectors
- move query closer towards those vectors by
- adding vectors for relevant documents to the query vector
- subtract vectors for the non-relevant documents to the query vector
variations
- add positive weights to relevant terms
- add negative weights to words found in non-relevant document
- remove terms that only appear in non-relevant document
relevance feedback performance
positive feedback is more valuable than negative feedback
penalise, not remove
why is relevance feedback seldom used
users generally reluctant to provide explicit feedback
long queries
types of user feedback
explicit
- relevance feedback
- query suggestion
- similar pages
implicit
- click tracking
- mouse tracking
- user behaviour
pseudo relevance feedback
assume the top retrieved documents are relevant and do rocchio method/ automatic query expansion
need to reweight the query terms
might drift away from optimal query
eg. homepage of glasgow information retrieval has only 1 unique document, not useful to expand query
thesaurus based query expansion
for each term in query, expand with related words and synonyms
weight the added terms less than original