In the meeting we talked about the latest situation of our TUBITAK report. Currently we finalized our tests and proceeding on report writing progress. We also talked about the paper: Hearst, M.A.: Texttiling: Segmenting text into multi-paragraph subtopic passages.
Computational Linguistics 23(1) (1997) 33-64
At the end of the meeting we took the following decisions:
In the meeting we revised Anil’s research documentation which is a new approach to search result clustering and labeling.
At the end of the meeting we took the following decisions:
In the meeting we did rehearsal of our plagiarism method (P^2CD) presentation. We also talked about Bilge’s diversification process.
At the end of the meeting we took the following decisions:
In the meeting we talked about latest situation of our external plagiarism and parallel corpora detection algorithm P^2CD. We also talked about details of language model implementation within the scope of information retrieval systems.
For the next week, we took the following decisions:
In the meeting we talked about our plagiarism detection algorithm (P^2CD) and its experimental results. The algorithm gives competitive results over the raw plagiarism corpus. Paired t-test results show that Levenstein metric has a better overall result for this dataset. However, our method is bilingual and it is not only designed for plagiarism detection but also for parallel corpora detection which can not be accomplished by using Levenstein distance since it is based on monolingual string comparisons.
In the meeting we talked about Bilge’s diversification research progress as well as our plagiarism detection method and Anil’s summarization process. According to test results, our external and plagiarism detection algorithm seems only working in non-obfuscated plagiarism cases (raw plagiarism).
For the next meeting we took the following decisions:
-Proceeding on tests over our plagiarism and parallel corpora detection algorithm.
In the meeting we talked about latest situation of our external plagiarism and parallel corpora detection algorithm. According to results, our method seems successful when the plagiarism doesn’t contain obfuscation. We also talked about Anil’s summarization process and Bilge’s diversification research progress.
At the end of the meeting we took the following decisions:
We talked in the meeting about the latest status of plagiarism detection algorithm as we as Anil’s summarization process. Currently near duplicate news detection with our plagiarism detection algorithm with considering sliding window size is working too slow.
At the end of the meeting we took the following decisions:
In the meeting we talked about Anil’s summarization process and our proposed external plagiarism and parallel corpora detection algorithm. We started using the plagiarism detection algorithm over our Bilkent Information Retrieval Group near duplicate dataset.
Currently the algorithm gives poor results over this dataset. This is because of the high level of obfuscation, ignorance of stepsize while adapting our algorithm for near duplicate dataset and inadequate ground truth information.
For the next meeting we took the following decisions:
We talked about the latest progress in our plagiarism detection approach. Currently we detected the best configuration over a test dataset and according to results, best configuration is detected as blocksize: 300 words, stepsize: 3 words, documentsize: 10 words, clusterthreshold (difference percentage between the number of clusters of compared blocks): 20%, yao difference threshold (difference between the actual cluster distribution average and yao (random) distribution): 20, consecutiveness threshold: 2, and pair threshold: 3.
At the end of the meeting we took the following decisions:
