• Visual information influences
• Cordeiro, Felicia, et al. "Rethinking the mobile food journal: Exploring opportunities for lightweight photo-based capture."
  Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems. ACM, 2015.
• Same type restaurants looks similar

• Content-based recommendation
• Lops, Pasquale, Marco De Gemmis, and Giovanni Semeraro. "Content-based recommender systems: State of the art and trends."
  Recommender systems handbook. Springer US, 2011. 73-105.
• Collaborative filtering recommendation
• Su, Xiaoyuan, and Taghi M. Khoshgoftaar. "A survey of collaborative filtering techniques."
  Advances in artificial intelligence 2009 (2009): 4.

• Visual properties aided hotel rating
• Chu, Wei-Ta, and Wei-Han Huang. "Cultural difference and visual information on hotel rating prediction."
  World Wide Web (2016): 1-25.

• Restaurant-dedicated social platform
• Restaurant names/ ID
• Restaurant: number of associate blog articles
• User ID/ pocket information (favorite collection)
• Texts and photos in blog articles


↑
Data Source

• Seamlessly and finely integrate visual information
   into the recommendation framework
• Combine visual features with text information

• Hybrid recommendation aided by visual information
• Combine content-based recommendation and
   collaborative filtering

• Text information
• Manually picked 2,118 related words
• 2,118 dimension feature vector based on blog articles

$\mathbf{a} = (a_1, a_2, ... , a_{2118})$

• User $u$ collect restaurants $R_u^+$ has preference $\mathbf{Ft_u}$

$\mathbf{Ft_u} = \sum_{i=1}^{m} \mathbf{a}^{(i)}$

• Visual information
• Convolutional Neural Network
  (CNN) features

Pre-trained vgg-f from
"MatConvNet Toolbox" / SVM / Max-pooling

• Conventional features
  - color name

14 primary colors ← 14-dimension vector

Framework of data preparation

Enhanced FM

Enhanced MF or BPRMF

$\mathbf{Sim}_{u_2,r_3}^a = \frac{\mathbf{P}_2^a \cdot \mathbf{v}_3}{\|\mathbf{P}_2^a\|\|\mathbf{v}_3\|}$

$\begin{eqnarray}\mathbf{P}_{i,j} = \begin{cases} 1, & \mathbf{Sim}_{u_2,r_3}^a \geq \psi \cr 0, & \mathbf{Sim}_{u_2,r_3}^a < \psi \end{cases} \end{eqnarray} $
and original $\mathbf{P}_{i,j} = 0$

• IEEE 2017 ICWS Special Track
• Due Deadline Changed... to 1 May
• Baselines:

  Finished: Data preparation
  Finished: Algorithm implementation

• WISE 2017 (Due Apr. 16) - project
• Finished: Objective function and solution
• Not Finished: Implementation
• Not Finished: Data preparation
• Not Finished: Paper writing

• Plan for this week
• WISE 2017

   Algorithm implementation
   Data preparation
   Start experiment
   draft writing