is there somebody from LucidWorks who can refer to Click Score Relevance Framework in LucidWorks Search?
On Mon, Apr 28, 2014 at 10:48 PM, nolim <alony...@gmail.com> wrote: > Hi, > We are using solr in production system for around ~500 users and we have > around ~10000 queries per day. > Our user's search topics most of the time static and repeat themselves over > time. > > We have in our system an option to specify "specific search subject" (we > also call it "specific information need") and most of our users are using > this option. > We keep in our system logs each query and document retrieved from each > "information need" > and the user can also give feedback if the document is relevant for his > "information need". > > We also have special query expansion technique and diversity algorithm > based > on MMR. > > We want to use this information from logs as data set for training our > ranking system > and preforming "Learning To Rank" for each "information need" or cluster of > "information needs". > We also want to give the user the option filter by "relevant" and "read" > based on his actions\friends actions in the same topic. > When he runs a query again or similar one he can skip already read > documents. That's an important requirement to our users. > > We think about 2 possibilities to implement it: > 1. Updating each item in solr and creating 2 fields named: "read", > "relevant". > Each field is multivalue field with the corresponding label of the > "information need". > When the user reads a document an update is sent to solr and the field > "read" gets a label with > the "information need" the user is working on... > Will cause update when each item is read by user (still nothing compare to > new items coming in each day). > We are saving information that "belongs" to the application in solr which > may be wrong architecture. > > 2. Save the information In DB, and then preforming filtering on the > retrieved results. > this option is much more complicated (We now have "fields" that aren't solr > and the user uses them for search). We won't get facets, autocomplete and > other nice stuff that a regular field in solr can have. > cost in preformances, we can''t retrieve easy: "give me top 10 documents > that answer the query and unread from the information need" and more > complicated code to hold. > > 3. Do you have more ideas? > > Which of those options is the better? > > Thanks in advance! > > > > -- > View this message in context: > http://lucene.472066.n3.nabble.com/saving-user-actions-on-item-in-solr-for-later-retrieval-tp4133558.html > Sent from the Solr - User mailing list archive at Nabble.com. > -- Sincerely yours Mikhail Khludnev Principal Engineer, Grid Dynamics <http://www.griddynamics.com> <mkhlud...@griddynamics.com>
