2. 对于每一个点类型对，有且仅有一个k步层次图与之对应；3. 融合了

2. For each point type pair, there is one and only one k-step hierarchical graph corresponding to it; <br>3. All the shortest element paths within k steps are merged, and the repeated parts in all meta paths are removed, reducing redundancy Yu; <br>4. This graphic structure not only discards a large amount of redundancy in the meta-heterogeneous information network, but also retains the partial graph information of the input data. It greatly reduces the size of the originally huge heterogeneous information network and facilitates path search. <br><br>We can build this structure from a large-scale heterogeneous information network with rich patterns, while the traditional network model is difficult to achieve. <br>On this type of data structure, the meta-path mining algorithm on the heterogeneous information network can be directly run without considering other issues such as type conversion. <br>At the same time for this structure, the starting node type is located in the center of the structure. All meta-path traversal algorithms starting from the starting node type, only when the generated meta-path does not end at the target node, or the length exceeds the dynamic length, Will exceed the boundaries of this structure. <br>At this time, the length of the meta path is often too long to contain effective information, so it can be ignored directly. <br>Therefore, the k-step pattern diagram designed in this paper can effectively replace the role of the network pattern in the meta-path mining algorithm. At the same time, because of this structure, the search space of the meta-path mining algorithm is greatly reduced, and the time efficiency of such algorithms is improved.

2. For each point type pair, there is and only one k-step hierarchy corresponding to it;<br>3. The shortest metapath within all k steps is fused, while removing the duplicate parts of all metapaths, reducing redundancy;<br>4. This graphical structure not only discards a large amount of redundancy in the metadata network, but also retains the local graph information of the input data. It greatly reduces the size of the original large heterogeneity information network and facilitates path search.<br><br>We can build this structure from large-scale, pattern-rich heterogeneity information network, while the traditional network model is difficult to achieve.<br>On such data structures, metapath mining algorithms on heterogeneic information networks can be run directly, without considering other problems such as type conversion.<br>At the same time for this structure, the starting node type is at the center of the structure, and all metapath traversal algorithms starting with the start node type will exceed the boundaries of the structure only if the resulting metapath does not end with the target node, or if the length exceeds the dynamic length.<br>At this time, the metapath length is often too long to contain valid information, so it can be directly ignored.<br>Therefore, the k-step pattern diagram designed in this paper can effectively replace the role of network pattern in metapath mining algorithm, and at the same time, because this structure greatly reduces the search space of metapath mining algorithm, and improves the time efficiency of such algorithm.

2. For each point type pair, there is and only one k-step hierarchical graph corresponding to it;<br>3. All the shortest meta paths within K steps are fused, and the redundant parts are removed;<br>4. This graph structure not only abandons a lot of redundancy in meta heterogeneous information network, but also retains the local graph information of input data. It greatly reduces the scale of the original huge heterogeneous information network and facilitates the path search.<br>We can construct this structure from a large-scale heterogeneous information network with rich patterns, but the traditional network model is difficult to achieve.<br>In this kind of data structure, we can directly run the meta path mining algorithm on heterogeneous information network without considering other problems such as type conversion.<br>At the same time, for this structure, the starting node type is located in the center of the structure. All the meta path traversal algorithms starting from the starting node type will exceed the boundary of the structure only when the generated meta path does not end at the target node or the length exceeds the dynamic length.<br>However, the length of meta path is often too long to contain effective information, so it can be ignored directly.<br>Therefore, the k-step pattern graph designed in this paper can effectively replace the role of network pattern in meta path mining algorithm. At the same time, because of this structure, the search space of meta path mining algorithm is greatly reduced, and the time efficiency of this kind of algorithm is improved.