Capturing the context / intention of a query

[brettasmi]

During a recent TRAPI call, we discussed the topic of how to explicitly capture a user's intent when creating a query. Many components can tweak their behavior contextually: an ARA may rank results contextually or a KP might respond with edges that are specific to a disease context.

The TRAPI issue focused on the implementation of a parameter in the query graph that could specify this context, but it was suggested that this needs to be tackled with the data modeling group and potentially others as well. Some notes on the discussion:

• This arose from an imProving Agent request to enable it to rank results contextually. Currently, it guesses context based on the nodes in the query graph, but it's clear that there are users who will formulate a query graph without encoding the context (disease, gene, substance, etc) as a node in the query graph itself.

• @edeutsch brought up the idea that KPs can work contextually and that there was some work on "context" in the data modeling group

• @mbrush confirmed that the data modeling group had considered this to some extent, but it was a lower priority for them at this point in time

• It was discussed that qualifiers / constraints might be sufficient to handle the KP contextual behavior

[sandrine-m]

MolePro user engagement work has a list of queries that could be used as examples to determine context strategies. During the past relay, we worked on identifying issues that could lead to different modeling strategies. As an example, we found that many of our questions were defining a baseline more or less implicitely:

1. What are druggable targets of [gene =?] specifically expressed in [tissue =?]?

Here the researcher is implicitely stating that the baseline must be all tissues but not the tissue of interest by using the adjective specifically.

2. What are genetic and proteomic correlates of a significant increase in [protein=?] activity in the extracellular matrix during E12-E16 development of lung alveoli branching?

Here the researcher is implicitely stating that the baseline must be at a temporal period outside of E12-E16. The researcher is only interested in correlating expression to the protein of interest only if it is present at 12 week to 16 week prenatal.