Text suggestions hackaton

This repository holds a small hands-on session to explore few ways of implementing text suggestions using Solr as a store for them. Some points:

• The aim is not using Solr as a whole for the text suggestions generation and ranking, but use it as a token/shingle/text store and let the inception of the suggestions happen outside.
  • We'll mostly use Solr to provide a blend of prefixed infixed suggestions.
• The aim would be to build these suggestions whenever this repository is downloaded, and executed through either Solr web UI or CLI, regardless how they were generated.
  • There would be some exceptions for some examples but mostly we'll use Solr.
• You should be familiar (or willing to play) with at least a bit of Python, a bit of Java, and a bit of Scala

---

Examples

1. Location-based text suggestions with latitude/longitude modelling and Geohashes [⚙️Work in progress, mostly done]
2. "Relevance" based iterations with supply/demand-based logic
3. Using Point-wise Mutual Information (PMI) index to generate next autocomplete token on text suggestions [⚙️Work in progress, generation but no indexation]

---

Requisites

• docker
• docker-compose
• wget
• make command

What comes with this repository

On the docker-compose.yml we have

• A solr9 local instance
• A mostly-fully-fledged jupyter environment with a Python kernel, whose main packages are
• An AWS Glue containerized environment to execute some examples.
  • Docker image is quite heavy so it is for the moment commented out from docker-compose for other examples

The idea for this structure is that we are providing the building blocks for suggestions with other tools (being them a jupyter environment or local AWS Glue) rather than plain Solr and we index them once they are ready.

For the early probers

Please execute on your terminal

 $ make down && make up && sleep 15 && make build_suggestions_1