To measure the latent similarity among documents, we construct topic vectors with the topic probabilistic distribution, and then adopt the Jensen-Shannon divergence to measures it, on the other hand, in the case of using document vectors we adopt cosine similarity .

Table 1: Extracting important sentences Methods Measure Accuracy F-value PageRank J enshen-Shannon 0.567 0.485 Cosine similarity 0.287 0.291 tf.

idf J enshen-Shannon 0.550 0.43 5 Cosine similarity 0.275 0.270

We computed the similarity between co-occurrence vectors using different metrics: Cosine Similarity , Dice coefficient (Curran, 2004), Kullback—Leibler divergence or KL divergence or relative entropy (Kullback and Leibler, 1951) and the J enson-Shannon divergence (Lee, 1999).

One year data (1991) were used to extract the “noun wo verb” tuples to compute word similarity (using cosine similarity metric) and collocation scores.

These data are necessary to compute the word similarity (using cosine similarity metric) and collocation scores.

The data was generated by clustering similar news stories from Gigaword using TF-IDF cosine similarity of their headlines.

Doc-pair Cosine Similarity

The x-axis represents the cosine similarity between the document pairs.

Additionally, there is more data in the MTC dataset which has low cosine similarity than in RF.

We can then estimate the topical similarity of two article sets by calculating the cosine similarity of their category frequency vectors C712=Aand6722= Bas

Table 3: Cosine similarity scores between the category frequency vectors of the flawed article sets and the respective random or reliable negatives

Borrowing this idea, for each sub-summary in a human-generated summary, we find its most matched sub-summary (judged by the cosine similarity measure) in the corresponding system-generated summary and then define the correlation according to the concordance between the two

For the semantic-based approach, we compare three different approaches to defining the subtopic number K: (1) Semantic-based 1: Following the approach proposed in (Li et al., 2007), we first derive the matrix of tweet cosine similarity .

For a tweet in a peak area, the linear combination of two measures is considered to evaluate its significance to be a sub-summary: (l) subtopic representativeness measured by the cosine similarity between the tweet and the centroid of all the tweets in the same peak area; (2) crowding endorsement measured by the times that the tweet is re-tweeted normalized by the total number of re-tweeting.

Evaluation: The similarity between a tweet and a news article is measured by cosine similarity .

4 The cosine similarity

In the WTMF model, we would like the latent vectors of two text nodes Q.,j1,Q.,j-2 to be as similar as possible, namely that their cosine similarity to be close to 1.

o 31(mi, mj): The cosine similarity of 75071;) and t(mj); and tweets are represented as TF-IDF vectors;

0 32(mi, mj): The cosine similarity of 75071;) and t(mj); and tweets are represented as topic distribution vectors;

SemTag uses the TAP knowledge base5, and employs the cosine similarity with TF-IDF weighting scheme to compute the match degree between a mention and an entity, achieving an accuracy of around 82%.