1.7.3 relevance-Correlation

1. Relevance

Correlation is a degree (degree) of a query response that satisfies the user's search information.

The relevance of the query response depends primarily on the context of the query. a single search application can be used in different contexts through the different needs and expectations of the user . For example , the search engine for climate data may be used by university researchers who have long studied the climate, and may also be cared for by spring The last frost of the day is used by farmers, and may also be used by civil engineers concerned with the frequency of rainfall patterns and floods , or it can be a college student program to an area that wants to know the climate of the area. as the motivations of these users change , so does the relevance of query responses.

　　How should the query response be synthesized? Like general relevance, the answer to this question depends on the scope of the search. In some cases, The cost of finding a specified document in a query response is high. For example, a legitimate e-discovery search response subpoena, search other time is slow, such as search cake recipes on the site of dozens of Or hundreds of cake recipes . When you configure SOLR, you should combine tradeoffs against other factors such as timeliness and ease of use.

Two important concepts of relevance:

• Precision (precision) : Returns the result, The percentage of the document's relevance.
• Recall : is the percentage of results that are returned from all relevant results, get Perfect recall very simple : Only each document in the collection is returned for each query.

back to the example above, we need a 100% recall (recall) All documents associated with the subpoena. This is important for e-discovery search applications, but it is less important to provide accuracy for a cake search application. However, in some cases, returning too many results in a casual context can overwhelm the user's perception. in some cases , return have correlation higher likelihood < Span lang= "ZH-CN" >< Span id= "Result_box" class= "Short_text" lang= "ZH-CN" > less Results Might be the best way to .

Using the concepts of precision and recall, it is possible to quantify the correlation between the user and the query for the document in the collection. A perfect system that will have 100% accuracy and a 100% recall (recall) for every user and query. In other words, It will retrieve all the relevant documents, nothing else. In the actual terms, it is common to focus on accuracy and recall in a number of results when discussing the accuracy and recall in a real system. The most common and useful results count when 10.

With facets, query filtering, and other search components, the SOLR app can be flexibly configured to adjust the search order to return the most relevant results for the user. In other words, SOLR can configure a balance of accuracy and recall (recall) to meet the needs of a specific user group.

　　 The configuration of the SOLR application should take into account :

• The needs of the various users of the application ( in addition to strict information requirements , can also include ease of use and response speed ,)
• span> Categories that are meaningful to users in a variety of environments. for example , date, product category, or region
• span> Span id= "Result_box" class= "Short_text" lang= "ZH-CN" > document any intrinsic associativity. For example, it might be meaningful to make sure that in the official product description or answer question always The returns the top of the search results for .
• span> Span class= "Short_text" lang= "ZH-CN" > The age of the document is an important flag.

< span> Consider All factors, it's in a SOLR deployment Planning Phase often Help outline the response type, you think Search application should return sample query . Once this is applied, and run. you can use a series of detection methods.

span> If you are interested in grouping, internal testing.

For more relevant information, refer to grant Ingersoll's technical documentation debugging Search application relevance issues.

1.7.3 relevance-Correlation