Training | To reduce computation, we employ NCE, which uses randomly sampled sentences from all target language sentences in Q as e‘, and calculate the expected values by a beam search with beam width W to truncate alignments with low scores. |
Training | where 6+ is a target language sentence aligned to f+ in the training data, i.e., (f+, 6+) 6 T, e‘ is a randomly sampled pseudo-target language sentence with length |e+|, and N denotes the number of pseudo-target language sentences per source sentence f+. |
Training | In a simple implementation, each 6— is generated by repeating a random sampling from a set of target words (V6) |e+| times and lining them up sequentially. |
Experiments | For the test data, we randomly sampled 23,650 examples of (event causality candidate, original sentence) among which 3,645 were positive from 2,451,254 event causality candidates extracted from our web corpus (Section 3.1). |
Experiments | Note that, for the diversity of the sampled scenarios, our sampling proceeded as follows: (i) Randomly sample a beginning event phrase from the generated scenarios. |
Experiments | (ii) Randomly sample an effect phrase for the beginning event phrase from the scenarios. |
Conclusions and Future Work | This may be an effect of ‘sparseness’ of relevant user data, in that users talk about politics very sporadically compared to a random sample of their neighbors. |
Identifying Twitter Social Graph | In the Fall of 2012, leading up to the elections, we randomly sampled n = 516 Democratic and m = 515 Republican users. |
Identifying Twitter Social Graph | For each such user we collect recent tweets and randomly sample their immediate k = 10 neighbors from follower, friend, user mention, reply, retweet and hashtag social circles. |
Identifying Twitter Social Graph | Similar to the candidate-centric graph, for each user we collect recent tweets and randomly sample user social circles in the Fall of 2012. |
Phase 1: Inducing the Page Taxonomy | Taxonomy quality To evaluate the quality of our page taxonomy we randomly sampled 1,000 Wikipedia pages. |
Phase 1: Inducing the Page Taxonomy | It was established by selecting the combination, among all possible permutations, which maximized precision on a tuning set of 100 randomly sampled pages, disjoint from our page dataset. |
Phase 3: Category taxonomy refinement | Category taxonomy quality To estimate the quality of the category taxonomy, we randomly sampled 1,000 categories and, for each of them, we manually associated the super-categories which were deemed to be appropriate hypemyms. |
Conclusions | Through manual evaluation we found that the algorithm could correctly identify 60.4% birth cases from a set of 48 random samples and 57% split/join cases from a set of 21 randomly picked samples. |
Evaluation framework | We selected 48 random samples of candidate words for birth cases and 21 random samples for split/join cases. |
Evaluation framework | A further analysis of the words marked due to birth in the random samples indicates that there are 22 technology-related words, 2 slangs, 3 economics related words and 2 general words. |
Data preparation | The parallel corpus is randomly sampled into two large and equally-sized parts. |
Data preparation | The final test set is created by randomly sampling the desired number of test instances. |
Experiments & Results | The final test sets are a randomly sampled 5, 000 sentence pairs from the 200, 000-sentence test split for each language pair. |