10.07.2015 Views

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

62 3 Algorithms <strong>and</strong> <strong>Techniques</strong>Here we discuss a special model, named homogeneous model. The word “homogeneous”means that the matrix T <strong>and</strong> the φ are always the same for the model no matter what kind ofconditional <strong>and</strong> emission distributions. As such, we can obtain the joint probability distributionover both latent variables <strong>and</strong> observations as[ N∏] N∏p(X,Y|θ)=p(y 1 |π) p(y n |y n−1 ,T) p(x m |y m ,φ). (3.10)n=2m=1where X = {x 1 ,...,x N }, Y = {y 1 ,...,y N }, <strong>and</strong> θ = {π,T,φ} denotes the set of parameterscontrolling the model. Note that the emission probabilities could be any possible type, e.g.,Gaussians, mixtures of Gaussians, <strong>and</strong> neural networks, since the emission density p(x|y) canbe directly obtained or indirectly got by using Bayes’ theorem based on the models.There are many variants of the st<strong>and</strong>ard HMM model, e.g., left-to-right HMM [211],generalized HMM [142], generalized pair HMM [195], <strong>and</strong> so forth. To estimate the parametersof an HMM, the most commonly used algorithm is named Expectation Maximization(EM) [72]. Several algorithms have b<strong>ee</strong>n later proposed to further improve the efficiency(or training HMM), such as f orward-backward algorithm [211], <strong>and</strong> sum-product approach.Readers can refer to [211, 33, 182] for more details.3.6 K-Nearest-NeighboringK-Nearest-Neighbor (kNN) approach is the most often used recommendation scoring algorithmin many recommender systems, which is to compare the current user activity with thehistoric records of other users for finding the top k users who share the most similar behaviorsto the current one. In conventional recommender systems, finding k nearest neighbors isusually accomplished by measuring the similarity in rating of items or visiting on web pagesbetw<strong>ee</strong>n current user <strong>and</strong> others. The found neighboring users are then used to produce a predictionof items that are potentially rated or visited but not done yet by the current active uservia collaborative filtering approaches. Therefore, the core component of the kNN algorithmis the similarity function that is used to measure the similarity or correlation betw<strong>ee</strong>n usersin terms of attribute vectors, in which each user activity is characterized as a sequence ofattributes associated with corresponding weights.3.7 Content-based RecommendationContent-based recommendation is a textual information filtering approach based on users historicratings on items. In a content-based recommendation, a user is associated with the attributesof the items that rated, <strong>and</strong> a user profile is learned from the attributes of the itemsto model the interest of the user. The recommendation score is computed by measuring thesimilarity of the attributes the user rated with those of not being rated, to determine whichattributes might be potentially rated by the same user. As a result of attribute similarity comparison,this method is actually a conventional information processing approach in the case ofrecommendation. The learned user profile reflects the long-time preference of a user withina period, <strong>and</strong> could be updated as more different rated attributes representing users interestare observed. Content-based recommendation is helpful for predicting individuals preferencesince it is on a basis of referring to the individuals historic rating data rather than taking otherspreference into consideration.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!