trep / opentrep

Open Travel Request Parser
https://trep.github.io/opentrep
GNU Lesser General Public License v2.1
11 stars 5 forks source link

Indexing travel types with PageRank weights #5

Open da115115 opened 10 years ago

da115115 commented 10 years ago

When someone searches for common types and/or countries, there is first a full text matching process, retrieving a maximum number of POR, for instance currently 30. Out of that limited list, the one with the highest PageRank value is then returned. That works well for cities (e.g., "paris airport"), but not so well for countries and/or travel types (e.g., "france airport", "uk railway", or just "usa" or "airport").

So, when indexing on the country level and/or travel type, it would be good to weigh with the PageRank value of the POR.

da115115 commented 10 years ago

The commits tracked by the [Weights] tag implement that requested feature:

The following requests now work as expected:

da115115 commented 10 years ago

Reopened because of the note below.

For some reasons (probably because the PageRank-based indexing weights are not big enough), the following requests bring not so bad answers, but not exactly what would be expected:

da115115 commented 10 years ago

See also #3 (Airport/city codes should take the precedence over alternate names) and #4 (Standsted STN).