bi-graph with Map-Reduce implementation

instance: $$(u,i,r_{ui},t_{ui})$$
describes user $u$ applies item $i$ on time $t_{ui}$ with score $r_{ui}$.

formular:
$$\begin{gathered} S_{ij}=\frac{1}{k_j^{\lambda }}\sum _u\frac{r_{ui}{r}_{uj}}{k_u^{\rho }}g(t_{ui},t_{uj}) \\ g(t_1,t_2)=\exp [-\frac{(t_1-t_2{)}^2}{2{\tau }^2}] \end{gathered}$$

parameters: $\lambda ,\rho ,\tau$

source data:
$$(u,p,r,t)$$

1. MAP:
   $$p:(u,r,t)$$
   REDUCE:
   $$p:[(u,r,t),(),...]$$
   calc: $k_p=(\sum r{)}^{\lambda }$,
   $$u:(p,r,t,k_p)$$
2. REDUCE:
   $$u:[(p,r,t,k_p),(),...]$$
3. Map
   calc: $k_u=(\sum r{)}^{\rho }$,
   p 0 : { ( p i , s i ) , ⋯   , } p_0: \{(p_i, s_i), \cdots,\} p0​:{(pi​,si​),⋯,}
   with : $s_i\to \frac{r_0{r}_i}{k_u{k}_{pi}}g(t_0-t_i)$

Reduce:
$$p_0:[(p_i,s_i),(),...]$$
with: $s_i\to {\sum }_{j=i}{s}_j$

Note

map(3)可合并到Reduce(2)中，但会极大增加2结果文件的大小，且会略微增加总耗时
时间对比：

#合并前
job1 time: 115s    1.9m
job2 time: 110s    1.83m
job3 time: 4551s   75.8m
total time: 4776s  79.6m

# 合并后
job1 time: 110s    1.83m
job2 time: 1217s   20.3m
job3 time: 3656s   60.9m
total time: 4983s  83m