[mythtv] Feature Request

Robert Middleswarth robert at middleswarth.net
Tue Mar 4 19:43:24 EST 2003


I think it could be done but would require alot of work.  Work I not 
sure anyone whats to put in.  You could take those shows that you have 
set to record automatical and use that as a starting point.  Then you 
would have to have the system learn making sugestions then based on your 
thumbs Up / Down / Notsure refine you sugestion even more.  It would 
take time and the sugestions would get better over time at 1st they 
would be off base at best.  Althought I agress that better descriptions 
would make it more accurate faster.  Also well could combine those 
though some form of Option P2P system system that would compare thing 
like show title and see what else someone else is also watch to help the 
data along over time.

Robert

Andy Davidoff wrote:

>There is no support for multiple genres in the database.  We could
>add this easily, though I'm not sure how useful it would be without
>our own addition of categorization hints.  In my tests[0], the "missing"
>categories are things like
>	"Adventure" from "Action/Adventure"
>or	"Fitness" from "Health/Fitness"
>or	"Tabloid" from "Talk/Tabloid"
>
>It's not clear that storing multiple genres really buys us anything
>without a better quality data source[0].
>
>I agree that the Naive Bayes is a good place to start if we're going
>to pursue statistical NLP, but I really doubt it can be effective on
>this problem without some sort of additional training data (thesaurus?).
>Our corpus is pretty small and the sentence fragments we do have are
>relatively short and not very descriptive or specific.
>
>I think getting actor information would really go a long way toward
>improving performance of suggestions, but where can we get such data?
>
>[0] This only applies to my data from tv_grab_na.
>
>
>#if Robert Middleswarth /* Mar 04, 18:25 */
>  
>
>>Does Myth have genre info coming in thought XMLTV?  The Best I could 
>>think we could do is base it off the summarys for the shows.  Use a 
>>method like a  Naive Bayes to teach myth what you like and don't like.
>>    
>>
>#endif /* robert at middleswarth.net */
>
>  
>



More information about the mythtv-dev mailing list