|
Posted by Booted Cat on October 7, 2006, 9:35 pm
Please log in for more thread options
By the way, I mentioned Microsoft's search macro idea weeks before
Microsoft debuted it in March 8, 2006, in this newsgroup thread. lol!
http://groups-beta.google.com/group/list.linguist/browse_frm/thread/d414eadca6631324/ba6d3c9be051bdb1?lnk=gst&q=%22search+command%22&rnum=1#ba6d3c9be051bdb1
But as you can see, I mentioned MORE THAN what Microsoft did.
> Below I describe the complete feature set that the search macro should
> have, including some that Microsoft already implemented.
>
> The basic idea is the use of "macros" in contrast of "raw keywords" to
> better represent concepts and predicates. Macros have these advantages:
>
> 1. Polysemy resolution. For example, if a user initially searches for:
> [ Apple ], System could suggest two macros which the user can use to
> refine his search criteria: @Apple_Fruit, @Apple_Inc. If @Apple_Fruit
> is used in his new search expression, this macro would actually be a
> raw keyword "apple" plus an OR expression of context words that include
> or exclude desired/undesired contexts, for example: apple AND (fruit
> OR food)
>
> 2. Synonymous matching. For example, if a user initially searches for [
> NLP ], System could suggest a macro @NLP whose actual definition is
> "NLP" OR "natural language processing" OR "computational linguistics".
>
> 3. Formal representation of relations (predicates). A relation (or
> "predicate") is a concept who takes at least one "argument" (therefore
> a concept is a 0-argument predicate). For example, if the user wants to
> find definitions of "domestication", there can be several surface
> patterns for this purpose: "the definition of ...", "... is defined
> as", "define ... as", "defines ... as", "defined ... as", along with
> other synonymous patterns. If System could suggest to the user a
> one-argument macro "@define(your_term)", all these common patterns can
> be encapsulated into this macro and the end user would only need to
> rewrite his search using this single macro.
>
> 4. Intelligent narrowing/broadening. A macro can have multiple versions
> of definitions, from the most preferred to the least preferred. A
> preferred version is supposed to return the best results, but can cause
> a risk of returning too few results. So, System should make a best
> compromise in choosing which version for each macro used in the user
> search expression. Some macros have to be "relaxed" to a less preferred
> version when necessary. Kind of like solving a linear equation system,
> isn't it?
>
> 5. Domain-specific macro suggestion. Based on an existing raw keyword
> or macro the user has typed into the search box, System could
> prompt/predict other domain-specific macros related to this user-given
> keyword/macro. The suggested macros could be organized as a dynamically
> generated ontology consisting of macros as nodes and their descriptions
> and descriptive arcs between the nodes.
>
> Microsoft's Windows Live Search already implemented 0-argument search
> macros in early March 2006, and recently introduced a taxonomy
> (http://gallery.live.com/default.aspx?l=4) for the users to pick up a
> desired macro. But two critical new features have to be added: (1)
> argument-taking macros (predicates) as in point #3 I showed above; (2)
> suggest related macros as the user types raw keywords in the search
> box, which is a practical approach to retrieve macros automatically.
>
> Regards,
> Yao Ziyuan
| 
XML Sitemap