Mostly, in recommendation system/collaborative filtering, cosine similarity approach is used. But, this metric is the angle between the two vectors and does not consider the magnitude of the values. Users (rating) on other hand, dont conider the angle between the vector. For example: [1 1 10 9] and [10 10 100 90] are more/exactly similar while [1 1 10 9] and [1 2 9 10] are less similar than the previous comparison. Which may lead to incorrectly result.
Dear Brother,
Cosine similarity is most commonly used for measuring similarity between two feature vectors. It can be used for both content based and collaborative filtering recommender systems. However, cosine similarity work well for content based recommender systems where we have to use TF-IDF weighting scheme to represent different items and documents. As far as the said issue is concerned, it mostly arises in collaborative filtering recommender systems. To cope with this problem an improved version of cosine similarity i.e., adjusted cosine similarity and Pearson correlation metrics can be used. Some researchers recommend to use a hybridization approach to handle this issue. However, shortcoming are still there.
Regards
Zafar Ali
Zafer Ali sir, Gulam Sulwar sir can we apply cosine similarity to find the traffic similarity in networking for different environments. Whether the parameters will be same or different. If any suggestion or such papers are present please provide it or give the reference of those.
Dear Sethi,
It will be better to use an enhanced similarity approach as identified by Zafar Ali to avoid the raised issue. You can use similarity anywhere. The temporal, QoS, QoE parameters may be used to find the similarity of the traffic flowing on the interfaces in the environment.
Regards,
Ghulam Sarwar sir, if you have any paper that provides the similarity measurement using cosine for network traffic analysis plz attach it.
Regards
one of our paper uses the network parameters in the similarity measure....
"A Network and Visual Quality Aware N-Screen Content Recommender System Using Joint Matrix Factorization", which may guide you to incorporate network/device/content parameters in the similarity and recommendation process.
- Badges
- Science topic