If you are an ML/NLP engineer, and share our inner passion to push boundaries of Conversational AI, or simply want to make NLP/NLU tools easier to use by the wider community, we welcome you to join our Open Source Community.
Go-Bot is a framework for building simple goal-oriented bots in DeepPavlov. Originally it supported training only with the DSTC2 Schema format. While it is quite rich and powerful, converting your data into DSTC-2 Schema might be a challenging task. Fortunately, one of our Open Source community members, Eugene, stepped in and built his own DSTC2-compliant dataset generation tool. After internal analysis and code review, we've been happy to release it as a part of our v0.11 release!
Top N Answers In ODQA Model
ODQA (open-domain question-answering) is one of the most popular components of DeepPavlov. It allows the computer to find answers directly in the raw text. However, initially you could only use this model to get the best answer from the candidates found . This made usage of the model rather limited . Thanks to Mapryl's Top N Answers PR to the ODQA model, now you can get N candidate answers with scores . It was shipped as part of our v0.12 release.
Relation Extraction
In release v0.17.0 we introduced a new relation extraction model based on the Adaptive Thresholding and Localized Context Pooling. Currently, RE is available for both English and Russian languages. You can learn more about supported relations and other things in the official documentation. In the English RE model, the output is Wikidata relation id and English relation name. In the Russian RE model, there is also an additional Russian relation name if it is available. This model was developed by one of our brave GSoC 2021 students, Anastasiia Sedova.
DD-IDDE project
Dmitry Babadeev became main contributor to our DD-IDDE project. DD-IDDE (a codename) stands for Discourse-Driven Integrated Dialog Development Environment. Dmitry jumped in, and became an owner of the VS Code extension that represents the "IDE" part of the DD-IDDE. Dmitry re-worked our original code that made conversion from Draw.io diagrams into DFF, as well as written conversion from DFF back to Draw.io. Dmitry made both conversions working offline within the extension as the original conversion from Draw.io to DFF DSL worked as an online web service.