论文阅读——Discovering Subsequence Patterns for Next POI Recommendation
00 文章基本信息
01 摘要
以往方法存在的不足：
当前主流方法在学习用户的check-in序列中的POI-level POI级（如何理解POI-level:比如一个学生一年到头都在学校里不出校门，他的所有的POIs无非就是教室、宿舍、食堂、图书馆，这样的单个POI连起来很难判断出用户的行为模式（有点不太好解释））序列时忽略了那些代表用户的社会经济活动和用户偏好一致性的子序列模式 （下面介绍什么是子序列模式）。但是由于很难定义那些complex but meaningful的子序列的granularity粒度 ，因此很难整个整合这些序列的语义子序列。
-
ignore the subsequence patterns that often represent the
socio-economic activities or coherence of preference of the users. -
However, it is challenging to integrate the semantic subsequences due to the difficulty to predefine the granularity of the complex but meaningful subsequences.
our contributions：
一个模型：ASPPA
- we propose Adaptive Sequence Partitioner with Power-law Attention (ASPPA) to automatically identify each semantic subsequence of POIs and discover their sequential patterns.
两种方法：a state-based stacked recurrent neural network 和a power-law attention mechanism
- a state-based stacked recurrent neural network to hierarchically learn the latent structures of the user’s check-in sequence.
- design a power-law attention mechanism to integrate the domain knowledge in spatial and temporal contexts.
1 Introduction
next POI recommandation的定义：
- next POI recommendation aims at predicting the POIs that are most likely to be visited next given the user’s check-in history.
next POI recommandation的不同之处：
- Compared with the general POI recommendation, next POI recommendation focuses more on exploiting user movement patterns hidden in the historical check-in sequence.
子序列模式：（很重要！）
这样理解：
首先什么是子序列模式：functional zones of city(cities)–>movement patterns of users–>check-in sequence of users（结合下图理解）
第二，由check-in sequence推出用户的行为模式。
①仅仅在一个城市功能区functional zone可能看不出用户的行为模式（比如一个学生一年到头都在学校里不出校门，他的所有的POIs无非就是教室、宿舍、食堂、图书馆，这样的单个POI连起来很难判断出用户的行为模式（有点不太好解释））；
②用户的行为模式极有可能只能通过用户在各个功能区（也就是很多POIs）之间的活动序列才能发现，而不是单个POI；
③此时就可以将一个功能区内的所有用户的check-in活动（多个POIs）看成一个整体whole。
本文待解决的两个问题：
- 第一，确定语义子序列identify the semantic subsequences并找出序列模式discover their sequential pattern；
- 第二，解决数据稀疏性问题；
最后总结本文的contibutions：（总结的很好）
-
第一，一个模型（ASPPA）实现两个任务（identify the semantic subsequences和discover their sequential patterns）： We design and develop a novel recommendation framework ASPPA to identify the semantic subsequences of POIs and discover their sequential patterns.
-
第二，提出一个机制（Power-law Attention mechanism）并解决一个问题（alleviates the data
sparsity and largely reduces down the number of parameters）： We propose a Power-law Attention mechanism to integrate the domain knowledge in both spatial and temporal contexts for the user’s check-in sequence. Our design alleviates the data sparsity and largely reduces down the number of parameters. -
第三，实验验证： We conduct extensive experiments to evaluate our framework on two widely used real-world datasets. The results show that our model is effective and outperform state-of-the-art techniques.
2 Related Work
3 Proposed Model
本文提出的模型叫做ASPPA：Adaptive Sequence Partitioner with Power-law Attention 。
顾名思义，其由两部分组成：Adaptive Sequence PartitionerASP和Power-law AttentionPA。
3.1 Problem Statement
几个定义如下：
-
用户集：U
-
POI集：L
-
用户的check-in history：
其中，each tuple (l t u , τ t ) is the t th visit of user u ∈ U , with
POI l t u ∈ L and time stamp τ t. -
next POI recommandation的问题定义：
The problem of next POI recommendation is to predict the POIs for user u at the next time step i.e. for each user we learn a personalized ranking function f u : V → Ω that maps the user’s check-in history to a probability distribution for the target POI set ：
The output is the top-k ranked POIs according to the predicted
probabilities.