Getting started with Nutritional Information
While there’s no single large feature that has landed in RecipeRadar in the past fortnight, there are a number of irons in the fire.
This week’s update is primarily textual rather than visual – we don’t have any screenshots to share this time – but as you’ll see, that’s appropriate given some of the work we’ve been doing.
The items we’re making progress on currently include:
Nutritional Information
We’ve identified a number of trustworthy and authoritative sources of ingredient nutritional information, including the USDA’s FoodData Central and UK’s CoFID dataset.
The next challenge is to find a good way to associate the records from these datasets with the ingredient identifiers that RecipeRadar uses internally, and to update the knowledge-graph so that it can store and serve the associated metadata.
Please note that we don’t intend to surface this nutritional information to users via the application during Q3; we do however have plans to offer functionality related to dietary planning in Q4, and that will incorporate nutritional factors.
Direction Metadata
Over the past couple of weeks, we’ve explored various options related to extraction of metadata from recipe text – we want to identify the various actions involved when preparing recipes ("combine the flour and milk") and also to discover the important entities involved in each step (“pre-heat the oven to 200 F”).
The ideal outcome of this – which could take some time – is that we will be able to use the co-references discovered in recipe directions to represent those visually, as in the example diagram included in the application’s vision statement.
As part of the work towards this, we have re-structured some of the responses of the knowledge-graph so that they return data in XML format. The primary driving reasons for this are that XML can represent data as a tree/graph (which we want) and also that it can represent entities interleaved in-line with natural language text (which we also want).
In addition, we are now performing basic identification of verbs in recipe directions. This is relatively straightforward thanks to modern open-source natural language processing libraries. We’ll need to introduce more advanced techniques from these libraries in order to reach a better understanding and representation of recipe text.
Coming up next
It’s hard to say precisely where we’ll get to with each of these goals within the next two weeks – perhaps moreso because our entire engineering team (currently a euphemism for ‘me’) will be on holiday until mid-late August.
Integrating nutritional data before then seems like an ambitious but achievable goal, and we may send a mini-update if we manage that.
Either way, you can expect the next ‘fortnightly’ update roughly a month from now.