Hi folks 🙂 Does anyone have familiarity with leadsemantics.com and their textDistil product? I have two use cases, one from my “real” job, and one as part of my PhD work. I’m wondering how good their solution is for converting text to knowledge-graph. Thanks!
They presented a their solution (textdistil) along with Fluree datastore today (actually few hours before).
Their ontology is built to map entities from specific domain (rental contracts).
Their classifier serves the purpose of entity extractions (& value) from raw texts which are mapped into the ontology built.
Unless you are thinking of using their current ontology, the Fluree Data store or the frontend pipeline. I don’t see a big advantage there considering that its a paid service.
Thanks Shashishekar R. - I was there 😉. that’s why I’m asking
I was thinking of using my own ontology. I was interested in the text to ontology NLP pipeline. I’ve tried to roll my own, but it’s not that easy - I mean, I’m getting poor results.
Hi Kyle This is Prasad from Lead Semantics, I presented 'TextDistil and Fluree' usecase in the webinar you mentioned. Thanks for attending the webinar. To clarify, the Ontology is specific to the domain and the application for which the extraction is meant for. Another domain and application would call for another Ontology. (we are not doing open domain extraction - that is not our focus, and there are many solutions to do that!) TextDistil is designed to work with different Ontologies. If you have an Ontology that is great, then with TextDistil, you only have to map the rest of the pipeline to fit your Ontology (and of course factor in implications of the input text corpus). In case I am not clear, please feel free to reach out to me.
Thanks Prasad Y.. Nice presentation. Are you guys thinking of opening an instance of TextDistil for the forum users to try ?. It would help to build and test individual ontologies.
Thanks and Thank you for attending the webinar Shashishekar R.. At this time, we have given access to one academic institute and educating ourselves on the cost implications. We love to give access to interested folk who want to try in a non-commercial setting. Hope to have an answer soon. Thanks for the suggestion.
Hi Prasad Y., I would be also interested in trying out TextDistil for research purposes. Together with colleagues of mine, we are trying to create a KG from free text about ancient Greek Mythology, based on a custom ontology we have created. Here is our first attempt (paper under review). The solution we are currently deploying does some very good NLP, but the ontology population is based on completely customised code. Hopefully TextDistil can make our lives easier. And ofc proper citations and acknowledgments will be given in due time if the solution fits our purposes. Many thanks in advance.
Hi Prasad Y. Thank you so much for your presentation and for your answer. My ontology is not quite ready for prime-time 😳. What I’m working on is modeling propaganda in news articles. So I’ve looked at some existing ontologies for rhetoric and arguments, but this is a pretty hairy beast. Based on the literature, this use case is best suited to building the ontology bottom up. So it’s a bit of a circular process as between building the ontology and the knowledge graph.
Hi Efstratios K. Thanks for the interest in TextDistil. Congratulations on the work and the paper. Thanks for sharing, reading the paper has given us some understanding of the goals (though, we have to reread portions of the paper closely!). TextDistil, while a reasonable architectural solution, I expect (code) changes be made in the rules and classifier modules (not just new training!). We have to check out CLTK (may be others to account for the archaic/extra-ordinary constructions in the text) how it can substitute/supplement the current lexical modules in the pipeline. At the other end of the pipeline - we are not yet clear about the support for reification statements for such things as contradictions ! I believe our team will be required for code changes and to help with (re) configuration along side you all, should we attempt it. Given the research (non revenue) nature of the project, it would be a matter of scheduling on our side. We are very much interested in the domain and open to discuss
