Automatically extracting keyphrases that are salient to the document meanings is an essential step to semantic document understanding. An effective keyphrase extraction (KPE) system can benefit a wide range of natural language processing and information retrieval tasks. Recent neural methods formulate the task as a document-to-keyphrase sequence-to-sequence task. These seq2seq learning models have shown promising results compared to previous KPE systems The recent progress in neural KPE is mostly observed in documents originating from the scientific domain. In real-world scenarios, most potential applications of KPE deal with diverse documents originating from sparse sources. These documents are unlikely to include the structure, prose and be as well written as scientific papers. They often include a much diverse document structure and reside in various domains whose contents target much wider audiences than scientists. To encourage the research community to develop a powerful neural model with key phrase extraction on open domains we have created OpenKP: a dataset of over 150,000 documents with the most relevant keyphrases generated by expert annotation.
There are important files that Microsoft projects should all have that are not present in this repository. A pull request has been opened to add the missing file(s). When the pr is merged this issue will be closed automatically.
Hi, when we're doing the case study using the evaluation script, we got some confusion about the results.
We find that you have the normalize_answer function. Several rules are set to normalize the candidates and references. However, we found these methods are not applied to the answers: in the return statement, these separate rules are not even called. This caused at least the punctuation problem: no punctuations are included in your raw text field of the dataset, but some punctuations exist in your keyphrases which makes it impossible for the models to predict the "exactly matched keyphrases".
Refer to:
We notice that you filter keyphrases beginning with an empty string "". However, there're some(quite a few) keypharses in the dataset which are not exactly empty like ["", "Middle", "East", "&", "Jewish", "World"] in line 6109 of dev.jsonl. So is there a reason to discard these kind of keyphrases?
Refer to:
I appreciate your effort in creating this great dataset and maintaining the leaderboard. Since the leaderboard for OpenKP has been retired, I wonder if there is any plan releasing the test dataset, so that the research community can continue studies with it.