WO2020114109A1

WO2020114109A1 - Interpretation method and apparatus for embedding result

Info

Publication number: WO2020114109A1
Application number: PCT/CN2019/112106
Authority: WO
Inventors: 张晓露; 王力; 向彪; 周俊
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2018-12-04
Filing date: 2019-10-21
Publication date: 2020-06-11
Also published as: TWI711934B; TW202022641A; CN109902167B; CN109902167A

Abstract

Disclosed are an interpretation method and apparatus for an embedding result. The method comprises: using an embedding algorithm to carry out embedding processing on embedded objects, and obtaining an embedding result of each embedded object, wherein the embedding result comprises embedding values under several dimensions (S102); according to an extreme value of the embedding values, extracting an embedded object of which an embedding value in each dimension satisfies a significant condition, and taking same as a significant training sample (S104); for each dimension, using a sample feature and significant category tag of the significant training sample under the dimension to train an interpretation model (106); and based on the trained interpretation model, determining an interpretation feature of the significant training sample belonging to a significant category, and taking same as an interpretation feature of the embedding result under the dimension (S108).

Description

Interpretation method and device of embedded result

Technical field

This specification relates to the field of machine learning technology, and in particular to an interpretation method and device for embedded results.

Background technique

Embedding represents a kind of mapping in mathematics, which can map one space to another space, and retain the basic attributes. The embedding algorithm can transform some complex and difficult-to-express features into easy-to-calculate forms, such as vectors and matrices, which are convenient for the prediction model to process. However, the embedding algorithm is not explanatory and cannot meet the needs of business scenarios.

Summary of the invention

In view of this, this specification provides a method and device for interpreting embedded results.

Specifically, this specification is implemented through the following technical solutions:

An interpretation method for embedded results, including:

Use an embedding algorithm to embed the embedded object to obtain the embedding result of each embedding object, the embedding result includes embedding values of several dimensions;

According to the extreme value of the embedding value, extract the embedding object whose embedding value meets the salient condition in each dimension as a salient training sample;

For each dimension, use the sample features and salient category labels of the salient training samples in this dimension to train the explanatory model;

Based on the trained interpretation model, it is determined that the salient training sample belongs to the interpretation feature of the salient category as the interpretation feature of the embedding result in this dimension.

A result interpretation method for graph embedding, including:

The embedding algorithm is used to embed the graph nodes to obtain the embedding result of each graph node, and the embedding result includes embedding values of several dimensions;

Extracting graph nodes whose embedding values meet the salient conditions in each dimension according to the extreme values of the embedding values as salient training samples;

A result interpretation method for word embedding, including:

The embedding algorithm is used to embed the vocabulary in the text to obtain the word embedding result corresponding to each text, and the word embedding result includes embedding values in several dimensions;

Extract the vocabulary whose embedding value meets the salient condition in each dimension according to the extreme value of the embedding value as a salient training sample;

A device for interpreting embedded results, including:

The embedding processing unit embeds the embedded objects by using an embedding algorithm to obtain the embedding result of each embedding object, and the embedding result includes embedding values of several dimensions;

The sample extraction unit extracts, as a significant training sample, an embedded object whose embedded value satisfies the significant condition in each dimension according to the extreme value of the embedded value;

The model training unit, for each dimension, uses the sample features of the salient training samples and salient category labels in that dimension to train the explanatory model;

The feature interpretation unit determines the interpretation feature of the salient training sample belonging to the salient category based on the trained interpretation model as the interpretation feature of the embedding result in this dimension.

A device for interpreting embedded results, including:

processor;

Memory for storing machine executable instructions;

Wherein, by reading and executing the machine executable instructions stored in the memory corresponding to the interpretation logic of the embedded result, the processor is prompted to:

As can be seen from the above description, this specification can extract embedded objects whose embedding value meets the salient conditions as salient training samples based on the extreme values of the embedding values in the embedding results as salient training samples, and use the salient training samples to interpret the interpretation. The model is trained, and then the interpretation characteristics of the embedded result in the corresponding dimension are determined according to the trained interpretation model, and the feature interpretation of the embedded result is realized, which provides a basis for the developer to repair the deviation of the original prediction model and helps improve the original prediction The generalization ability and performance of the model, and help to avoid legal risks and moral hazards.

BRIEF DESCRIPTION

FIG. 1 is a schematic flowchart of an embedding result interpretation method shown in an exemplary embodiment of this specification.

FIG. 2 is a schematic flowchart of another method for interpreting an embedded result shown in an exemplary embodiment of this specification.

FIG. 3 is a schematic structural diagram of an apparatus for interpreting embedded results shown in an exemplary embodiment of the present specification.

Fig. 4 is a block diagram of an apparatus for interpreting an embedded result shown in an exemplary embodiment of this specification.

detailed description

Exemplary embodiments will be described in detail here, examples of which are shown in the drawings. When referring to the drawings below, unless otherwise indicated, the same numerals in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.

The terminology used in this specification is for the purpose of describing particular embodiments only, and is not intended to limit this description. The singular forms "a", "said" and "the" used in this specification and the appended claims are also intended to include most forms unless the context clearly indicates other meanings. It should also be understood that the term "and/or" as used herein refers to and includes any or all possible combinations of one or more associated listed items.

It should be understood that although the terms first, second, third, etc. may be used to describe various information in this specification, the information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as second information, and similarly, the second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "when" or "in response to a determination".

This specification provides an interpretation solution for embedding results. An embedding algorithm can be used to embed an embedded object to obtain an embedding result that includes embedding values in several dimensions. Then, based on the extreme value of the embedding value, extract significant training samples in each dimension The training model is used to train the interpretation model, and the interpretation features of the significant training samples are used as the interpretation features of the embedding result in the corresponding dimension to realize the interpretation of the embedding result.

FIG. 1 and FIG. 2 are flowcharts of a method for explaining an embedding result shown in an exemplary embodiment of this specification.

The embedding algorithm may include a graph embedding (Graph Embedding) algorithm, and the graph embedding algorithm may map the graph data to low-dimensional dense embedding results, such as vectors, matrices, etc.; the embedding algorithm may also include: word embedding (Word Embedding) Algorithm, word embedding can map vocabulary into low-dimensional embedding results, such as vector, matrix, etc.

Please refer to FIG. 1 and FIG. 2, the method for interpreting the embedded result may include the following steps:

Step S102, embedding the embedded objects using an embedding algorithm to obtain an embedding result of each embedding object, the embedding result including embedding values of several dimensions.

In one example, the embedded object may be a graph node in the graph structure.

For example, the embedded object may be a user node in a user network graph. The user network map may be established based on the user's payment data, friend relationship data, and the like.

After the embedding algorithm is used to embed the user nodes in the user network graph, the vector corresponding to each user node can be obtained.

In another example, the embedded object may be text to be clustered, such as news, information, and the like.

The embedding algorithm is used to embed the vocabulary included in each text, and the vector corresponding to each vocabulary in each text can be obtained, and the vector set corresponding to each text can be obtained.

In this embodiment, for ease of description, vectors, matrices, etc. obtained by embedding objects processed by an embedding algorithm may be collectively referred to as embedding results. The embedding result may include embedding values in several dimensions.

When the embedding result is a vector, each element of the vector can be regarded as one dimension, and each element value is an embedding value in the corresponding dimension.

When the embedding result is a matrix, each element of the matrix may also be regarded as a dimension, and each element value is an embedding value in the corresponding dimension.

When the embedding result is a matrix, each row or column of the matrix can also be regarded as a dimension. Taking behavior as an example, each row of the matrix can be regarded as a row vector, and then the sum of squares of each element in the row vector can be used as the embedded value in the corresponding dimension. Of course, in other examples, the elements and values of the row vector or the mean value of the elements may also be used as the embedded values in the corresponding dimensions, which is not particularly limited in this specification.

In this embodiment, after an embedding algorithm is used to embed each embedding object to obtain an embedding result, the embedding results of different embedding objects include embedding values of the same dimension. The embedded value is usually a value in the real number space, and is not interpretable.

For example, assuming that there are 100 embedded objects, after embedding the embedded objects using an embedding algorithm, the resulting embedding result includes a 50-dimensional vector. In other words, the embedding result vector obtained after the embedding process has 50 elements. In the present embodiment, the m-th embedded object can be obtained after processing the embedded embedding result vector referred to as _{_{E m, E m = {e}} m1, e m2, ..., e m50}.

In this embodiment, after the embedding result of each embedding object is obtained, the extreme value of all embedding values can be obtained.

In one example, the original prediction model may be trained using the embedding result of each embedded object, and after the training is completed, the original prediction model may output the extreme value of the embedding value in the embedding result.

For example, a storage bit may be added to the original prediction model to record the extreme value of the embedded value passing through the model network unit, and the extreme value may be output after the model is trained.

The above-mentioned original prediction models may include: classification models, regression models, clustering models, and so on.

In other examples, the extreme value of the embedded value may also be obtained in other ways, which is not specifically limited in this specification.

In this embodiment, the extreme value may include a maximum value and a minimum value. Still taking 100 embedded objects, the embedding result obtained by using the embedding algorithm includes the embedding value of 50 dimensions as an example. After training the original prediction model in this step, the maximum value of 5000 embedding values (100×50) can be obtained e _max and minimum value e _min .

In step S104, an embedding object whose embedding value meets the salient condition in each dimension is extracted as a salient training sample according to the extreme value of the embedding value.

In this embodiment, the saliency condition of the saliency training sample may be determined according to the extreme value of the embedding value, and then the embedding object whose embedding value meets the saliency condition in each dimension is extracted as the saliency training sample in the dimension.

In this embodiment, the extreme value includes a maximum value and a minimum value. Corresponding to extreme values, the salient conditions may include salient activation conditions and salient suppression conditions, the salient training samples may include salient activation training samples and salient suppression training samples, and the salient category label of the salient activation training samples is salient activation Label, the significant category label of the significant suppression training sample is a significant suppression label.

Wherein, the significant activation condition is that the embedding value is greater than or equal to the difference between the maximum value and the preset change parameter, and at the same time is less than or equal to the maximum value. Assuming that δ is used to represent the preset change parameter, the value range of the embedded value e _i that satisfies the significant activation condition is: e _max -δ≤e _i ≤e _max .

The significant suppression condition is that the embedding value is greater than or equal to the above minimum value, and at the same time is less than or equal to the sum value of the minimum value and the preset change parameter. That is, the value range of the embedding value e _i that satisfies the significant suppression condition is: e _min ≦e _i ≦e _min +δ.

In this embodiment, an embedded object that satisfies the aforementioned significant activation condition may be referred to as a significant activation training sample, and an embedded object that satisfies the aforementioned significant suppression condition may be referred to as a significant suppression training sample.

In this embodiment, after determining the salient activation condition and the salient suppression condition, for each dimension of the embedding result, salient activation training samples and salient suppression training samples can be extracted.

Taking the first dimension of the embedding result as an example, it can be judged in sequence whether the first embedding value of the embedding result of each embedding object processed by the embedding algorithm satisfies the above-mentioned significant suppression condition or significant activation condition. The embedded object is used as the salient training sample in the first dimension.

For example, please refer to the m-th embedded object in the foregoing step S102. In this step, it can be determined whether the first embedding value _em1 of the embedding result of this embedded object satisfies the aforementioned significant activation condition or significant suppression condition. If the above significant activation condition is satisfied, the embedded object can be extracted as the significant activation training sample in the first dimension; if the above significant suppression condition is satisfied, the embedded object can be extracted as the significant inhibition training sample in the first dimension; If neither is satisfied, it can be confirmed that the embedded object cannot be used as a significant training sample in the first dimension.

Similarly, for the second dimension of the embedding result, it can be judged in sequence whether the second embedding value of the embedding result of each embedding object satisfies the above significant suppression condition or significant activation condition, and if one of them is satisfied, the embedding object can be extracted as Significant training samples in the second dimension.

For example, the determination in step S102 is embedded in the m-th results of the embedded object embedded in the second value e _m2 meets the above conditions significantly or significantly inhibiting the activation conditions and the like.

In this embodiment, the same embedded object may be a significantly activated training sample in some dimensions, and may also be a significantly suppressed training sample in other dimensions.

For example, the embedded object m may be a significant activation training sample in the first dimension, a significant suppression training sample in the second dimension, and not a significant training sample in the third dimension.

In this embodiment, based on this step, significant training samples can be extracted for each dimension.

In step S106, for each dimension, a significant training sample in that dimension is used to train the interpretation model.

In this embodiment, the explanatory model may be a binary classification model with better interpretability, such as a linear model, a decision tree, etc., which is not particularly limited in this specification. It is worth noting that, since the multi-classification model is a special form of two-classification model, the above-mentioned two-classification model may include a multi-classification model.

In this embodiment, the interpretation model can be trained using the sample features and sample labels of the salient training samples.

Wherein, the sample label may be determined based on the previously trained prediction model.

The sample features may include original features and topological features of the sample.

The original feature is usually a feature already present in the sample itself.

For example, the original characteristics of the user node may include the user's age, gender, occupation, income, and so on.

As another example, the original features of the text may include part of speech of the vocabulary, word frequency, and so on.

The topological features can be used to represent the topological structure of the embedded object.

Taking an embedded object as a graph node as an example, the topological characteristics may include: first-order neighbor data, number of second-order neighbors, average number of neighbors of first-order neighbors, statistics of first-order neighbors under a specified original feature dimension, etc.

Taking risk gang identification as an example, the statistics of the first-order neighbors under the specified original feature dimension may be the average age of the first-order neighbors, the maximum age of the first-order neighbors, the average annual income of the first-order neighbors, and the Minimum annual income, etc.

Taking the embedded object as a vocabulary included in the text as an example, the topological features may include: the vocabulary that appears most often in front of the vocabulary, the number of vocabularies that often appear in conjunction with the vocabulary, and so on.

In this embodiment, topological features are used to supplement the original features. On the one hand, it can solve the problem that some samples have no original features. On the other hand, the topological structure of the samples can be added to the sample features, thereby improving the interpretation of the training results of the model. accuracy.

In this embodiment, for each dimension, after completing the training of the interpretation model, the weight of each sample feature in that dimension can be obtained.

Table 1

Please refer to the example in Table 1. In dimension 1, the weight of sample feature 1 is W11, the weight of sample feature 2 is W12...; in dimension 2, the weight of sample feature 1 is W21, and the weight of sample feature 2 is W22... Wait.

In step S108, the interpretation features of the significant training samples are determined based on the trained interpretation model as the interpretation features of the embedding result in this dimension.

Based on the foregoing step S106, the weight of each sample feature can be determined based on the interpreted model trained in each dimension, and according to the weight, several sample features that significantly affect the prediction result in the corresponding dimension can be determined as the interpreted features of the significant training sample. In this embodiment, the interpretation feature of the significant training sample may also be determined as the interpretation feature of the embedding result in this dimension.

For example, the sample features may be sorted according to the order of weight from large to small, and then the sample features arranged in the top N bits are extracted as the interpretation features. Among them, the value of N can be set in advance, N can be equal to 3, 5, etc., this specification does not make any special restrictions.

Please continue to refer to the example in Table 1. Assuming that in dimension 1, W11>W12>W13>Wi, and the value of N is 3, the interpreted features of the embedded result in dimension 1 can be determined as feature 1, feature 2 and feature 3.

This specification also provides a method for interpreting the results of graph embedding.

On the one hand, an embedding algorithm may be used to embed the graph nodes to obtain an embedding result of each graph node, and the embedding result includes embedding values in several dimensions.

On the other hand, the graph nodes whose embedding values meet the salient conditions in each dimension can be extracted as salient training samples according to the extreme values of the embedding values, and then for each dimension, the sample features and salient categories of the salient training samples in that dimension are used The label trains the interpretation model, and can determine the interpretation feature of the salient training sample belonging to the salient category based on the trained interpretation model as the interpretation feature of the embedding result in this dimension.

Taking a user network diagram as an example, in this embodiment, a user network diagram may be constructed based on data such as user payment data and interaction data. For each user node in the user network graph, an embedding algorithm can be used to obtain the embedding result of the user node, such as a vector.

According to the extreme value of the embedding value, user nodes whose embedding value meets the salient condition in each dimension can be extracted as a salient training sample.

For each dimension of each embedding result, the sample features and salient category labels of the salient training samples in this dimension can be used to train the interpretation model, and the interpretation features of the embedding result in this dimension can be determined based on the trained interpretation model .

For example, the interpretation characteristics of the embedding result in dimension 1 may include: no fixed occupation, annual income less than 80,000, resident place in Guangxi, age 18-25 years old, etc.

For another example, the interpretation characteristics of the embedding result under dimension 2 may include: no fixed occupation, annual income less than 100,000, place of residence in Yunnan, age 20-28 years old, SSID using Wi-Fi network is 12345, etc.

This specification also provides a method for interpreting the results of word embedding.

On the one hand, an embedding algorithm may be used to embed words in the text to obtain a word embedding result corresponding to each text, and the word embedding result includes embedding values in several dimensions.

On the other hand, according to the extreme value of the embedding value, the vocabulary whose embedding value meets the salient condition in each dimension can be extracted as a salient training sample, and then for each dimension, the sample features and salient category labels of the salient training sample in that dimension are used The interpretation model is trained, and the interpretation feature of the salient training sample belonging to the salient category can be determined based on the trained interpretation model as the interpretation feature of the embedding result in this dimension.

For example, the interpretation characteristics of the embedded result in dimension 1 may include: computer, artificial intelligence, technology, innovation, the word frequency of technology is greater than 0.01, etc.

As another example, the interpretation characteristics of the embedded result under dimension 2 may include: football, basketball, sports, swimming, recording, etc.

It should be noted that since a text usually includes several vocabularies, the word embedding result corresponding to the text may be a mosaic of the embedding results of each vocabulary included in the text, or each of the embedding results of each vocabulary. The embedding value is average added, etc., this manual does not make any special restrictions.

When extracting salient training samples, if the number of dimensions of the embedding result corresponding to the text is the same as the number of dimensions of the embedding result of the vocabulary, then the salient training samples can also be extracted in units of text, which is not particularly limited in this specification.

Corresponding to the foregoing embodiments of the embedding result interpretation method, this specification also provides an embodiment of the embedding result interpretation device.

The embodiment of the apparatus for interpreting embedded results in this specification can be applied to a server. The device embodiments may be implemented by software, or by hardware or a combination of hardware and software. Taking software implementation as an example, as a logical device, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory through the processor of the server where it is located and running. From the hardware level, as shown in Figure 3, this is a hardware structure diagram of the server where the interpretation device of the embedded result is located, except for the processor, memory, network interface, and non-volatile memory shown in Figure 3 In addition, in the embodiment, the server where the device is located usually includes other hardware according to the actual function of the server, which will not be repeated here.

Please refer to FIG. 4. The apparatus 300 for interpreting embedded results may be applied to the server shown in FIG. 3, including: an embedding processing unit 301, a sample extraction unit 302, a model training unit 303 and a feature interpretation unit 304.

The embedding processing unit 301 uses an embedding algorithm to embed the embedding objects to obtain embedding results of each embedding object, and the embedding results include embedding values of several dimensions;

The sample extraction unit 302 extracts embedded objects whose embedding values satisfy the salient conditions in each dimension according to the extreme values of the embedding values as salient training samples;

The model training unit 303, for each dimension, uses the sample features of the salient training samples in the dimension and the salient category labels to train the explanatory model;

The feature interpretation unit 304 determines the interpretation feature of the salient training sample belonging to the salient category based on the trained interpretation model as the interpretation feature of the embedding result in this dimension.

Optionally, the extreme value includes: a maximum value and a minimum value;

The significant conditions include: significant activation conditions and significant inhibition conditions;

The salient category tags correspond to the salient conditions, including salient activation tags and salient suppression tags;

The determination process of the salient condition includes:

Calculating the difference between the maximum value and the preset change parameter;

Calculating the sum of the minimum value and the preset change parameter;

Determining the significant activation condition as: the embedding value is greater than or equal to the difference and less than or equal to the maximum value;

The significant suppression condition is determined as: the embedding value is greater than or equal to the minimum value and less than or equal to the summation value.

Optionally, the feature interpretation unit 304:

Determine the weight of each sample feature in the significant training sample based on the trained explanatory model;

Sort the sample features according to the order of weight from large to small;

The sample features ranked in the top N are extracted as the explanatory features of the saliency training sample belonging to the saliency category, and N is a natural number greater than or equal to 1.

Optionally, the sample features include: original features and topological features.

Optionally, the topological features include one or more of the following:

The number of first-order neighbors, the number of second-order neighbors, the average number of first-order neighbors, and the statistics of first-order neighbors under the specified original feature dimensions.

Optionally, the interpretation model is a binary classification model.

For the implementation process of the functions and functions of the units in the above device, please refer to the implementation process of the corresponding steps in the above method for details, which will not be repeated here.

As for the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located One place, or can be distributed to multiple network elements. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solution in this specification. Those of ordinary skill in the art can understand and implement without paying creative labor.

The system, device, module or unit explained in the above embodiments may be specifically implemented by a computer chip or entity, or implemented by a product with a certain function. A typical implementation device is a computer, and the specific form of the computer may be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email sending and receiving device, and a game control Desk, tablet computer, wearable device, or any combination of these devices.

Corresponding to the foregoing embodiments of the method for interpreting embedded results, this specification also provides an apparatus for interpreting embedded results, which includes a processor and a memory for storing machine-executable instructions. Among them, the processor and the memory are usually connected to each other via an internal bus. In other possible implementations, the device may also include an external interface to be able to communicate with other devices or components.

In this embodiment, by reading and executing the machine-executable instructions stored in the memory corresponding to the interpretation logic of the embedded result, the processor is prompted to:

Optionally, the extreme value includes: a maximum value and a minimum value;

The determination process of the salient condition includes:

Calculating the sum of the minimum value and the preset change parameter;

Optionally, when it is determined that the saliency training sample belongs to the interpretation feature of the saliency category based on the trained interpretation model, the processor is prompted to:

Sort the sample features according to the order of weight from large to small;

Optionally, the topological features include one or more of the following:

Optionally, the interpretation model is a binary classification model.

Corresponding to the foregoing embodiment of the embedding result interpretation method, this specification also provides a computer-readable storage medium that stores a computer program on the computer-readable storage medium, and the program implements the following steps when executed by the processor:

Optionally, the extreme value includes: a maximum value and a minimum value;

The determination process of the salient condition includes:

Calculating the sum of the minimum value and the preset change parameter;

Optionally, the determining the interpretation feature of the salient training sample belonging to the salient category based on the trained interpretation model includes:

Sort the sample features according to the order of weight from large to small;

Optionally, the topological features include one or more of the following:

Optionally, the interpretation model is a binary classification model.

The foregoing describes specific embodiments of the present specification. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order shown or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The above are only the preferred embodiments of this specification and are not intended to limit this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of this specification should be included in this specification Within the scope of protection.

Claims

An interpretation method for embedded results, including:

Use an embedding algorithm to embed the embedded object to obtain the embedding result of each embedding object, the embedding result includes embedding values of several dimensions;

According to the extreme value of the embedding value, extract the embedding object whose embedding value meets the salient condition in each dimension as a salient training sample;

For each dimension, use the sample features and salient category labels of the salient training samples in this dimension to train the explanatory model;

Based on the trained interpretation model, it is determined that the salient training sample belongs to the interpretation feature of the salient category as the interpretation feature of the embedding result in this dimension.
The method according to claim 1,

The extreme value includes: a maximum value and a minimum value;

The significant conditions include: significant activation conditions and significant inhibition conditions;

The salient category tags correspond to the salient conditions, including salient activation tags and salient suppression tags;

The determination process of the salient condition includes:

Calculating the difference between the maximum value and the preset change parameter;

Calculating the sum of the minimum value and the preset change parameter;

Determining the significant activation condition as: the embedding value is greater than or equal to the difference and less than or equal to the maximum value;

The significant suppression condition is determined as: the embedding value is greater than or equal to the minimum value and less than or equal to the summation value.
The method according to claim 1, determining that the saliency training sample belongs to the saliency category interpretation feature based on the trained interpretation model, including:

Determine the weight of each sample feature in the significant training sample based on the trained explanatory model;

Sort the sample features according to the order of weight from large to small;

The sample features ranked in the top N are extracted as the explanatory features of the saliency training sample belonging to the saliency category, and N is a natural number greater than or equal to 1.
The method according to claim 3,

The sample features include: original features and topological features.
The method according to claim 4, the topological features include one or more of the following:

The number of first-order neighbors, the number of second-order neighbors, the average number of first-order neighbors, and the statistics of first-order neighbors under the specified original feature dimensions.
The method according to claim 1,

The explanation model is a binary classification model.
A result interpretation method for graph embedding, including:

The embedding algorithm is used to embed the graph nodes to obtain the embedding result of each graph node, and the embedding result includes embedding values of several dimensions;

Extracting graph nodes whose embedding values meet the salient conditions in each dimension according to the extreme values of the embedding values as salient training samples;

For each dimension, use the sample features and salient category labels of the salient training samples in this dimension to train the explanatory model;

Based on the trained interpretation model, it is determined that the salient training sample belongs to the interpretation feature of the salient category as the interpretation feature of the embedding result in this dimension.
A result interpretation method for word embedding, including:

The embedding algorithm is used to embed the vocabulary in the text to obtain the word embedding result corresponding to each text, and the word embedding result includes embedding values in several dimensions;

Extract the vocabulary whose embedding value meets the salient condition in each dimension according to the extreme value of the embedding value as a salient training sample;

For each dimension, use the sample features and salient category labels of the salient training samples in this dimension to train the explanatory model;

Based on the trained interpretation model, it is determined that the salient training sample belongs to the interpretation feature of the salient category as the interpretation feature of the embedding result in this dimension.
A device for interpreting embedded results, including:

The embedding processing unit embeds the embedded objects by using an embedding algorithm to obtain the embedding result of each embedding object, and the embedding result includes embedding values of several dimensions;

The sample extraction unit extracts, as a significant training sample, an embedded object whose embedded value satisfies the significant condition in each dimension according to the extreme value of the embedded value;

The model training unit, for each dimension, uses the sample features of the salient training samples and salient category labels in that dimension to train the explanatory model;

The feature interpretation unit determines the interpretation feature of the salient training sample belonging to the salient category based on the trained interpretation model as the interpretation feature of the embedding result in this dimension.
The device according to claim 9,

The extreme value includes: a maximum value and a minimum value;

The significant conditions include: significant activation conditions and significant inhibition conditions;

The salient category tags correspond to the salient conditions, including salient activation tags and salient suppression tags;

The determination process of the salient condition includes:

Calculating the difference between the maximum value and the preset change parameter;

Calculating the sum of the minimum value and the preset change parameter;

Determining the significant activation condition as: the embedding value is greater than or equal to the difference and less than or equal to the maximum value;

The significant suppression condition is determined as: the embedding value is greater than or equal to the minimum value and less than or equal to the summation value.
The apparatus according to claim 9, the feature interpretation unit:

Determine the weight of each sample feature in the significant training sample based on the trained explanatory model;

Sort the sample features according to the order of weight from large to small;

The sample features ranked in the top N are extracted as explanatory features of the saliency training sample belonging to the saliency category, and N is a natural number greater than or equal to 1.
The device according to claim 11,

The sample features include: original features and topological features.
The apparatus according to claim 12, the topological characteristics include one or more of the following:

The number of first-order neighbors, the number of second-order neighbors, the average number of first-order neighbors, and the statistics of first-order neighbors under the specified original feature dimensions.
The device according to claim 9,

The explanation model is a binary classification model.
A device for interpreting embedded results, including:

processor;

Memory for storing machine executable instructions;

Wherein, by reading and executing the machine executable instructions stored in the memory corresponding to the interpretation logic of the embedded result, the processor is prompted to:

Use an embedding algorithm to embed the embedded object to obtain the embedding result of each embedding object, the embedding result includes embedding values of several dimensions;

According to the extreme value of the embedding value, extract the embedding object whose embedding value meets the salient condition in each dimension as a salient training sample;

For each dimension, use the sample features and salient category labels of the salient training samples in this dimension to train the explanatory model;

Based on the trained interpretation model, it is determined that the salient training sample belongs to the interpretation feature of the salient category as the interpretation feature of the embedding result in this dimension.