En:Ranking

Definition of Ranking Rules

Ranking is the technical instance of relevance, which is 'what the user thinks is important'. Since almost every user has a different opinion about such 'relevance' it should be possible to define a personal ranking for a personal relevance. Also in different search environments the ranking should change, i.e. within a site-only search the user expects more in-depth results in contrast to popular documents in unlimited internet search.

A ranking can be done by the application of coefficients to document attributes. Such attributes can be various counts, like a count of words, bytes, outbound links, inbound links (from same or other domains) and so on. There is a large number of such attributes in YaCy. To see them all, open localhost:8090/IndexControlURLs_p.html in your YaCy peer. The ranking is done using Solr ranking methods. If your peer does a remote search, then also your own ranking configuration is applied to the remote peer. The configuration interface for this is in http://localhost:8090/RankingSolr_p.html

For a distributed search in YaCy there is a special challenge: some results from other peers may be returned without this rich set of attributes. They return a so-called Reverse Word Index References containing a small set of attributes. To combine search results from other peers, they must be ranked using this subset of attributes. We call that process 'post-ranking'. The configuration interface for this is in http://localhost:8090/RankingRWI_p.html. If you use YaCy for a portal search or intranet search without p2p mode, then you don't need to care about it.

Solr Ranking in /RankingSolr_p.html

The default ranking in Solr is done using 'boosts' on search fields. You can define which fields in the index Schema can be searched when you submit a search and how strong they are boosted on every field using the "Solr Boosts" section within /RankingSolr_p.html. If you would like to have search results, which have matches with your search word in a specific field higher, then increase the boost value for that specific field. Each field will produce a ranking value and the resulting ranking score will be the sum of these boosted fields multiplied with a norming factor which is the inverse of the square root of the sum of squared weights.

The boosted field ranking is just a formula for the relevancy of the search hit regarding the text content and construction of the document, but not for the topology of the document within the searched document set. To influence the ranking your can use Boost Queries and Boost Functions.

Boost Function

A Boost Function is a formula which evaluates attributes which are not search fields. You can use this i.e. to evaluate the date fields (for a date ordering) or to move documents with a lower clickdepth to the front. Example formulars are:

recip(ms(NOW,last_modified),3.16e-11,1,1)

to sort by last_modified and bring the most current documents to the top,

div(add(references_internal_i,product(references_external_i,references_exthosts_i)),add(clickdepth_i,1))

to consider documents with many external references as more important but also count the clickdepth. Just try to do your own experiments using the attributes from the index schema.

Boost Query

A Boost Query is a query that is attached to the other queries that YaCy calculates from the boost field configuration. It can be considered as a default, fixed add-on to a search. The purpose is, to boost specific attributes within the index schema. By default, we set this to

fuzzy_signature_unique_b:true^100000.0

Which boost a field which assignes that a document is unique. This moves double content to the back. You can use this like a default filter for your documents which move forward or backward in the result list independently from the searched word.

Filter Query

This is not actually a ranking mechanism, but certain use cases require that search results shall be 'boosted' to the end of the search result list (i.e. by boosting the reverse to the front) or should be removed completely from the results. To do so, a filter query can be assigned in this field to filter out unwanted results. The content of the field is a search term which must match with all wanted results. An example is the removal of all 'double' documents as flagged with the attributes http_unique_b and/or www_unique_b. If you do not want http documents if the same url with https is present and if you want to remove all documents without leading 'www.' if one with 'www.' is present, enter the following term:

http_unique_b=true AND www_unique_b=true

It is also possible to remove all documents with special words in the text part, i.e. if you want to remove all documents containing the word 'foo', use the following term:

-text_t:foo

Solr Boosts

Here you can find a list of all searchable text fields. Only such fields are actually used in a document search which are filled with a boost > 0 (empty boost fields mean boost = 0). The higher the boost value, the more relevant is the search field.