Hi Kenji, Find my answers inline.
On 3/9/15 10:56 AM, Kenji Yamauchi wrote: > Hello, everyone. > > Could you answer the following questions to clarify the project "Fact > Extraction from Wikipedia Text" and could you suggest the existing > warm-up tasks related to this project, if exist? I dropped a mail with the link, repasting it: https://github.com/dbpedia/fact-extractor > (I've chosen Relation Extraction as my master thesis topic and I would > like to combine it with this project, i.e. this project would be > practical part of my master thesis, If I could.) > > Firstly, just a confirmation, is it the aim of this project that we > enrich the existing datasets? For instance, given the thing in [1], do > we finally add new facts about the thing by using the developed > framework? Yes, you got it. > > Secondly, is it fixed that we use FrameNet for the fact extraction? > Could I use other approaches, such as Distant Supervision[2][3][4], to > extract facts if they are proper? I would prefer a Frame-based representation (as I'm confident with it), but I am completely open to distant supervision. So feel free to motivate and expand this, both here and in your proposal. Can you repaste reference 4, as I don't see it in the list below? > > Finally, if we use FrameNet, how about you consider following 3 points? > 1. As far as I understand, the project description shows only the flow > of extracting frames from the source text. How should I use the > frames? Do we directly use the frames as new relations between the > entities, or that we convert the Frames to existing DBpedia's RDF > properties (such as dbpedia-owl:successor) after the extraction? Good point. I would prioritize what emerges from the corpus, in a data-driven way, and model the frames accordingly. Whenever possible, we should reuse existing resources like FrameNet, and existing DBpedia (or other aligned schemas) ontology properties. In other words, this means: A. get some results first B. model them to fit into DBpedia > > 2. On the step "Verb extraction and ranking", which verbs is the > target of "ranking"? > All of the verbs in the article, as Peresa said? Or can I propose other > targets? All the domain-specific verbs that emerge from a weighted measure (i.e., TF-IDF) against the corpus. > > 3. Creating frames seems to depend on each language. Will I develop > the framework on English articles? Frames should be language-agnostic actually. However,we should start from a specific domain (i.e., soccer) in a specific language (i.e., English), and see if the learned frames can be generalized to all the languages DBpedia extracts. Cheers! > > Thank you and regards, > Kenji Yamauchi > > [1] http://www.dbpedia.org/page/It%C5%8D_Hirobumi > [2] http://nlp.stanford.edu/software/mimlre.shtml > [3] http://nlp.stanford.edu/pubs/emnlp2012-mimlre.pdf > > --------- > Kenji Yamauchi > Master's course on Kyoto University > > ------------------------------------------------------------------------------ > Dive into the World of Parallel Programming The Go Parallel Website, sponsored > by Intel and developed in partnership with Slashdot Media, is your hub for all > things parallel software development, from weekly thought leadership blogs to > news, videos, case studies, tutorials and more. Take a look and join the > conversation now. http://goparallel.sourceforge.net/ > _______________________________________________ > Dbpedia-gsoc mailing list > Dbpedia-gsoc@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/dbpedia-gsoc > -- Marco Fossati http://about.me/marco.fossati Twitter: @hjfocs Skype: hell_j ------------------------------------------------------------------------------ Dive into the World of Parallel Programming The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ Dbpedia-gsoc mailing list Dbpedia-gsoc@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dbpedia-gsoc
