Frequently Asked Questions (FAQ)¶
Why did I get fewer than 3 suggestions?¶
The LLM is asked for 3 suggestions, but the ones that are empty, duplicated, or identical to the query you sent are dropped. The list can therefore have fewer than 3 entries, or even be empty.
Why do I get different suggestions for the same query?¶
The suggestions are generated by an LLM, which is not deterministic. The user's keywords can also change between calls, as the User Profiler updates them.
Are the suggestions personalised?¶
Only if the request context contains a user_id known to the User Profiler, and the token you send is also accepted by the User Profiler. Otherwise, the suggestions depend on the current query only. See Security.
Does the service use the results of my query?¶
Not yet. The context object accepts any additional fields, but only user_id is used at the moment.
Does the service remember my previous queries?¶
No. The service is stateless. Your past behaviour is taken into account only through the interest keywords maintained by the User Profiler.
I get a 403 with "Token missing required audience"¶
Your token was issued for another client. Request a token whose audience includes query-recommender.
Known Issues¶
- A suggestion is not checked against the queries the user already made, so following the suggestions can lead back to a previous query.
- The LLM call has no timeout of its own: a slow LLM endpoint makes the request slow too, up to the timeout of the client or of the ingress.
GET /redirects to/api/v1without taking the proxy prefix into account. Behind the DataGEMS ingress, use/qr/api/v1directly.