19 lines
1.0 KiB
Markdown
19 lines
1.0 KiB
Markdown
## Question 1
|
|
- Term suggestion: suggest terms that split the query space.
|
|
- E.g., `jaguar`: add the word `car` or the word `cat`.
|
|
- Don't focus on adding similar terms -- limited utility.
|
|
- Want to suggest a diverse number of terms.
|
|
- We want to suggest terms that are maximally dissimilar to each other while still be similar to the original query.
|
|
- Trade-off: could maximise diversity by picking random terms, but these would not be relevant to the query.
|
|
- Want to suggest terms that make a more specific query.
|
|
|
|
## Question 2
|
|
- Term-term correlation: know co-occurrence of terms, e.g., t1 tends to occur with t2.
|
|
- User-user correlation: now that some users are making similar queries.
|
|
- Could suggest terms that similar searchers are using that maybe they haven't.
|
|
- Ignoring temporal evidence.
|
|
- Multiple term suggestions in ranked order preferable.
|
|
- Consider relation to autofill in Google.
|
|
- Impossible to get fully right, looking for identification of data being used and a valid approach to using it, plus advantages & disadvantages of this approach.
|
|
|