Tongji Data Structure Project - Social Network Links Prediction

video

• Name generator: Name generator
• organizations (schools)
  • high school
  • junior school
  • primary school

• javascript, python(+ jupyter notebook)
• echarts, bootstrap, ajax, networkx
• Django

create virtual environment in conda (env name: social_network)

cd network_demo
conda env create -f environment.yaml
conda activate social_network

An overview

enter the folder with manage.py then start service

cd network_demo
python manage.py runserver

scale background graph by scrolling click on categories at the top of the page

features implemented in echarts

click on cards in the left sidebar, profile displayed on the right sidebar

select Algorithms (default to Intersection) and Mode (default to All Users) at the bottom of Detailed infomation

All user mode

target user mode

wrong input get warnings

manage your page with control island (UDCP Buttons)

upload your JSON file at Custom

new graph loaded

return to Defalut graph at Default

data generator in folder data\eupho_dataset\dataset_generator.ipynb

As for the name eupho, I initially wanted to build up a network with the purpose of displaying sophisticated interpersonal relationships in Animation Series Sound! Euphonium. And that's exactly the reason why I tested Japanese names in my demo dataset. The organizations in this network could be schools, clubs, companies or even social groups and person nodes are individuals (characters in the anime :) )

Documenting the profiles of characters in Sound! Euphonium is really time-consuming which is not feasible, and actually unnecessary, for a 2-week project. Therefore, I got some random names from name generators on Google. (But I got names of Kituji High School Concert Band which are listed in the file data\eupho_dataset\dataset_generator.ipynb from Season 1 to Season 2.5)

$C(u, v)$ denotes the prediction score between individual $u$ and $v$.

$C(u, v) = \sum_{k \in \text{categories}} w_{u, k} \cdot N_{u, v, k}$

$C(u, v) = \frac{|\Gamma(u) \cap \Gamma(v)|}{|\Gamma(u) \cup \Gamma(v)|}$

where $\Gamma(u)$ denotes the set of neighbors of $u$, $\Gamma(v)$ denotes the set of neighbors of $v$

Bonus: individual social habits considered. A wide range of friends means less strengthed friendship while deeper friendship expected for individuals who have fewer connections with others. The node would earn higher score if it had a few connections but still shared many nodes with the target.

Official Networkx Implementation

$C(u, v) = \sum_{w \in \Gamma(u) \cap \Gamma(v)} \frac{1}{\log |\Gamma(w)|}$

where $\Gamma(u)$ denotes the set of neighbors of $u$. This index leads to zero-division for nodes only connected via self-loops. It is intended to be used when no self-loops are present. .

Official Networkx Implementation

$C(u, v) = \alpha \cdot (|\Gamma (u){\cap }^{}\Gamma (v)|)+(1-\alpha )\cdot \frac{N}{{d}_{uv}}, \quad \alpha \in [0, 1]$

where $\Gamma(u)$ denotes the set of neighbors of $u$, $\Gamma(v)$ denotes the set of neighbors of $v$, $\alpha$ is parameter varies between $[0, 1]$ and default to 0.8, $N$ denotes total number of nodes in the Graph and $d_{u, v}$ denotes shortest distance between $u$ and $v$ .