# Solid Practictioners Meeting (On Query)

Note : a video of this meeting is available at https://solid-practioners.trinpod.us/node/t_fa.

## Attendance
- [Jackson Morgan](https://o.team)
- [Noel De Martin](https://noeldemartin.com/)
- [Pierre-Antoine Champin](https://champin.net/#pa)
- Pete Edwards
- Tim Standen
- [michal/mrkvon](https://id.mrkvon.org) - sleepy.bike
- [Ruben Eschauzier](https://knows.idlab.ugent.be/)
- Helen Oliver - ESPRESSO
- Alain Bourgeois
- Alex Tourski
- Benoit A
- Els de Vleeschauwer - UGent/imec
- Fred Gibson (TrinPod, Graphmetrix)
- Jeff Zucker
- Maxime Lecoq
- Wilfred Pinfold
- Dmitri Z.
- Hadrian
- Talita A.
- Kai Gilb - Graphmetrix
- Mohamed Ragab - ESPRESSO
- Maarten Vandenbrande - UGent/imec
- Jesse
- James Doe Jr.

## Housekeeping
- JZ: The next solid world will be devoted to Solid Practitioners
- JZ: We're trying to do a hackathon about practitioners, if so there are any projects we should make the focus?

## Maxime Lecoq (Using Indexes to Search for Data in a Solid Ecosystem)
- ML: I am working with Fabien Gandon and Pierre-Antoine Champin in collaboration with Startinblox to build semantic applications
- ML: We want to find storage solutions that are as fast as possible.
- 2 Projects:
- Hubl: Find Freelancers
- Mycelium: Find agricultural products
- Browsing all resources would be very expensive, so we're using indexes
- Simple index can look like "A list of users where "A" is the first letter of their name"
- Federated or Distributed
- Federated: Many Pods indexed in one Pod
- Distributed: Each Pod has its own indexes
- How do we integrate them within Solid?
- We can add it in the client to server protocol
- Or use the client to client protocol (We'll explore this first as we don't need to change the client to server protocol)
- Client to client protocol defines ontologies for how to exchange data. The server doesn't need to know about it.
- Indexing can be done by application on the client side, either by the browser application or by bots..
- Framework to compare indexing strategies was built: made from a browser app and docker containers with Solid Community Server.
- Used to simulate a deployment of Startinblox
- User is indexed by 600 skills and 10 location cities
- Size of the index varies from 1 tiple to 150 triples
- A demo is shown of a web interface that allows the user to select and indexing strategy as well as insert a query. It shows what the backend query looks like. The query is launched showing the results as well as the execution time.
- Experiment Results:
- (Federated Query) Duration of query scales exponentionally based on the number of skills queried
- For distributed query, querying over 4 skills takes a lot of time.
- The same trend applies for the cities experiments
- Federated search executes in under 1.2 seconds for 5 skills and 5 cities
- In the future:
- Want to test strategies to index strings
- Exploid neamed graphs
- Create a query planner
- In the long terms we want to test more strategies and propose a vocabulary for generic indexing and discovery
- Contact at: [e-mail], @lecoqlibre on gitter

## Fred Gibson (Search Ideas for a new Solid API Specification)
- FG: We've developed our own serach API for TrinPod, and we'd love to have it implemented on Solid
- The actual API is `/search/`.
- Plain text, just give the text. Example: `https://pod.com/search/search%20term`
- Paging: `/search/term?limit=100`
- Much of this is inspired by Apache Solr
- Logic: "+" adds a term, "-" excludes a term
- The text that's searched in the Pod is any literal string
- Text search must be a direct match
- Tag: using an ontology tag:tag. Any subject with a `tag:tag` predicate will be tagged
- Type Search (Coming Soon): Any term that can be serialized as a uri can be searched.
- Pod Context: Assuming Pods are on the server you're talking to it will search. Selecting specific Pods has a query param.
- The response could be Triples as well as a rank from Apache Solr
- Also includes the version
- Additional context: heirarchy all the way up to the root.
- The ontology is also included in the response
- Demo: A search menu searches all three of the Pods on Trin Apps. It can do a text search of anything that has the word "profile." He puts the word "profile" in quotes and it only shows results that have "profile" with lowercase. He types `"/profile/"` and it shows objects in a folder. He then shows a request that constrains a search to a single Pod. For tag searches, he puts a `#` in front of a tag, and it returns objects with that tag.

## Helen Oliver (Efficient Search over Personal Repositories - Secure and Sovereign)
- HO: Yuri was supposed to present, but he wasn't able to make it. I'll only be able to give a shallow introduction to Espresso.
- JZ: We'd still love to hear from you.
- HO: I represent the Espresso Project. Involves the University of Southhampton and University of London.
- Link: https://espressoproject.org
- It's an experimental framework that carries out a distributed search over Solid pods.
- Takes varying access control into account.
- On each Pod server, we create an "Expresso Pod" and it goes through each server we have access to to index each file in the server.
- Brewmaster is a Pod indexing app. It relates to the meta-index in the expresso Pod as well as the inverted index of the content on the search app. We then use that metaindex to search over those Pods. We use Gaian DB to enable federated search.
- All things in Espresso:
- Pod Indexing App
- Pod Search App
- Overlay Network of Solid Servers
- Query User Interface
- If you have any questions, email to me and I'll pass them onto Yuri or Mohammed.
- Link to paper: https://link.springer.com/chapter/10.1007/978-981-99-7254-8_28

## Other Approaches
- Peter Edwards: Work with Inrupt, and wanted to use the Java SDK. Everything I'm doing is server side with a client built on top of it. We've created about 3000 indexes on the day's index so far. I store SPARQL queries in the same Pod.
- Alex Tourski: If we're talking about the number of Pods in the future. We're going to need to coordinate between existing Pods and legacy platforms like Facebook.
- Fred: Maybe we should also talk about search on Pods and Search that's managing everying (like google). Those are two different things. But, it would seem that if we have any search API.
- Noel De Martin: The way I solved search in my app is to have all the data locally. In an application where you only have personal data (which also plays really well with an offline-first approach, for the apps where that makes sense).
- Pierre-Antoine: There might be an in-between with the approach that you represent. The indexes could be used for general purpose, like type index.
- Mohammed: How do you bind access control with data?
- FG: We have a hypergraph that gives a URI to a triple. We have a patent pending for a linked data hypergraph. We have "attributes," "states," "events," and "processes," and each are URIs. We use Apache Solr.