Solr click scoring implementation

Question

after searching and searching over the net, i've found a possible open-source solution for the click-count-popularity in solr (=does not require a payd version of lucid work search).

In my next two answers i will try to solve the problem in a easy way and in a way a little bit complex...

But first some pre-requisites.

We suppose to google-like scenario:
1. the user will introduce some terms in a textfield and push the search button
2. the system (a custom web-app coupled with solr) will produce a web page with results that are clickable
3. the user will select one of the results (e.g. to access to the details) and will inform the system to change the 'popularity' of the selected result

Answer 1

The very easy way.

We define a field called 'popularity' in solr schema.xml

<field name="popularity" type="long" indexed="true" stored="true"/>

We suppose the user will click on the document with id 1234, so we (=the webapp) have to call solr to update the popularity field of the document with id 1234 using the url

http://mysolrappserver/solr/update?commit=true

and posting in the body

<add>
  <doc>
    <field name="id">**1234**</field>
    <field name="popularity" update="inc">1</field>
  </doc>
</add>

So, each time the webapp will query something to solr (combining/ordering the solr 'boost' field with our custom 'popularity' field) we will obtain a list ordered also by popularity

Answer 2

The more complex idea is to update the solr index tracing not only the user selection but also the search terms used to obtain the list.

First of all we have to define a history field where to store the search terms used:

<field name="searchHistory" type="text_general" stored="true" indexed="true" multiValued="true"/>

Then we suppose the user searched 'something' and selected from the result list the document with id 1234. The webapp will call the solr instance at the url

 http://mysolrappserver/solr/update?commit=true

adding a new value to the field searchHistory

<add>
  <doc>
    <field name="id">**1234**</field>
    <field name="searchHistory" update="add">**something**</field>
  </doc>
</add>

finally, using the solr termfreq function in every following query we will obtain a 'score' that combined with 'boost' field can produce a sorted list based of click-count-popularity (and the history of search terms).

Answer 3

This is interesting approach however I see some disadvantages in it:

1. Overall items storage will grow dramatically with each and every search.
2. You're assuming that choosing specific item is 100% correct and it wasn't done by mistake or for brief only. In this way you might get wrong search results along the way.

I suggest only to increment the counter or even to maintain relative counter based on the other results that the user didn't click it.