CN117689354B - Intelligent processing method and platform for recruitment information based on cloud service - Google Patents

Abstract

The application provides an intelligent processing method and platform for recruitment information based on cloud service, wherein the method comprises the steps that first position original information is sent to a recruitment information management platform through a first recruitment terminal, so that the recruitment information management platform generates first position description information according to the first position original information and a preset position description information generation model, and sends the first position description information to the first recruitment terminal, the first recruitment terminal responds to the first position description information, first position feedback information is sent to the recruitment information management platform, if the first position feedback information comprises first position release information, the recruitment information management platform releases the first position description information, generates first iteration training information according to the first position original information and the first position description information, and adds the first iteration training information to an iteration training database so as to conduct subsequent iteration training on a deep neural network model, and the purpose of effectively updating the model is achieved.

Description

Intelligent processing method and platform for recruitment information based on cloud service

Technical Field

The application relates to a data processing technology, in particular to an intelligent processing method and platform for recruitment information based on cloud service.

Background

With the rapid development of network technology, the network recruitment information platform provides efficient and convenient recruitment services for enterprises and job seekers, and is one of important means of human resource management.

The recruitment information platform can rapidly adapt to recruitment requirements of enterprises by utilizing the characteristics of high availability, elastic expansion and flexibility of cloud services, and provides stable and reliable services. Meanwhile, the application of the intelligent processing algorithm can reduce the burden of manual processing, improve recruitment efficiency and reduce cost.

However, the intelligent algorithm model based on the existing recruitment information platform is usually trained based on a preset fixed training sample, and after the training is completed, the iteration cannot be further effectively updated.

Disclosure of Invention

The application provides an intelligent processing method and platform for recruitment information based on cloud service, which are used for solving the technical problem that an existing intelligent algorithm model is usually trained based on a preset fixed training sample, and iteration cannot be further and effectively updated after training is completed.

The application provides an intelligent recruitment information processing method based on cloud service, which is applied to a recruitment information management system, wherein the recruitment information management system comprises a recruitment terminal set and a recruitment information management platform arranged on a cloud server, and the recruitment information management platform is in communication connection with each recruitment terminal in the recruitment terminal set; the method comprises the following steps:

A first recruitment terminal in the recruitment terminal set sends first post original information to the recruitment information management platform, wherein the first post original information comprises a first feature word set, and the first feature word set comprises a plurality of post feature words;

The recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to the first recruitment terminal, wherein the preset post description information generation model is a model generated based on training of a preset deep neural network model;

the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform;

If the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iteration training information according to the first post original information and the first post description information, and stores the first iteration training information as new iteration training information into an iteration training database corresponding to the preset deep neural network model, wherein the first iteration training information is configured with a first type of training feature labels.

Optionally, after the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform, the method further includes:

If the first post feedback information includes first post modification information and first post release information, the recruitment information management platform releases the first post modification information, and generates second iteration training information according to the first post original information, the first post description information and the first post modification information, so as to store the second iteration training information as new iteration training information into the iteration training database, wherein the second iteration training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information.

Optionally, after the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform, the method further includes:

If the first post feedback information comprises a first post modification request and the first post release information, the recruitment information management platform determines historical target post description information from a post release historical database according to the first feature word set, and determines a second recruitment terminal corresponding to the historical target post description information in the recruitment terminal set;

the recruitment information management platform sends first post correction information to the second recruitment terminal, wherein the first post correction information comprises the first post original information, the first post description information and the first post modification request;

The second recruitment terminal responds to the first post correction information and sends second post description information to the recruitment information management platform so that the recruitment information management platform sends the second post description information to the first recruitment terminal;

The first recruitment terminal responds to the second post description information and sends second post feedback information to the recruitment information management platform;

If the second post feedback information includes second post release information, the recruitment information management platform releases the second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information, so as to store the third iteration training information as new iteration training information into the iteration training database, wherein the third iteration training information is configured with the second class training feature tag, and the second post description information is post description information modified based on the first post description information.

Optionally, the recruitment information management platform determines, according to the first feature word set, historical target post description information from a post release history database, including:

The recruitment information management platform utilizes a Word2Vec Word embedding model and based on the first feature Word set Generating a first set of representations/>Wherein/>For the first feature word set/>/>Personal characteristic words,/>For the first feature word set/>/>Word embedding vector corresponding to each characteristic word,/>For the first feature word set/>The number of feature words in (a);

The recruitment information management platform utilizes jieba word segmentation tools and generates a first word segmentation set according to post release history information in the post release history database And determining the first Word segmentation set/>, by using Word2Vec Word embedding modelCorresponding second representation setWherein/>For the first word segment set/>/>Word segmentation,/>For the first word segment set/>/>Word embedding vector corresponding to each word segment,/>For the first word segment set/>The number of word segments in the database;

The recruitment information management platform calculates the first feature word set />Word embedding vector/>, corresponding to each feature wordAnd the second representation set/>Cosine similarity among the embedded vectors of each word;

If the word is embedded in the vector With word embedding vector/>Cosine similarity between/>If the recruitment information management platform is larger than the preset cosine similarity threshold, the recruitment information management platform determines the/>Individual feature words/>At the first word segmentation set/>The characteristic feature words exist in the Chinese character string;

The recruitment information management platform traverses the first set of representations Each word of the characteristic words is embedded into the vector, and if the quantity of the characteristic words is confirmed to exist/>, the characteristic words are representedAnd the first feature word set/>Number of feature words in/>The ratio of the post release historical information to the historical post description information to be selected is larger than a preset ratio threshold, and the historical post description information to be selected/>Wherein/>Candidate set/>, for the historical post description information/>Post release history information,/>Candidate set/>, for the historical post description informationTotal number of post release history information;

The recruitment information management platform calculates the first feature word set according to formula (1) And/>Post release history information/>Cosine similarity mean value between/>The formula (1) is:

（1）；

wherein, To at/>Post release history information/>The corresponding second representation set/>Chinese and word embedding vector/>The word with the highest cosine similarity is embedded into the vector;

the recruitment information management platform determines a maximum cosine similarity average value according to a formula (2) The formula (2) is:

（2）；

The recruitment information management platform determines the maximum cosine similarity average value The corresponding post release history information is the history target post description information.

Optionally, before the recruitment information management platform sends the first post correction information to the second recruitment terminal, the method further includes:

The recruitment information management platform determines the corresponding calling ratio of the published first post description information, wherein the calling ratio is the ratio between the delivery number of resume corresponding to the first post description information and the information browsing number;

and the recruitment information management platform determines that the calling ratio is smaller than a preset calling ratio threshold value.

Optionally, before the recruitment information management platform generates the first post description information according to the first post original information and the preset post description information generation model, the recruitment information management platform further includes:

creating an initial training database on the recruitment information management platform, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial characteristic word set and initial calibration post description information corresponding to the initial characteristic word set;

training the preset depth neural network model by using the initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder built on the basis of a cyclic neural network or a long-short-time memory network, and the sequence decoder is a decoder built on the basis of the cyclic neural network or an attention mechanism; wherein the step of each training comprises steps a to d:

step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

Step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of the preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at a minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate the preset post description information generation model.

Optionally, after storing the first iterative training information in an iterative training database corresponding to the preset deep neural network model, the method further includes:

The recruitment information management platform determines the number of newly-added information corresponding to newly-added iterative training information in the iterative training database, and if the number of newly-added information exceeds a preset information number threshold, and the ratio between the number of newly-added information of the second type corresponding to newly-added iterative training information configured with the second type training feature tag and the number of newly-added information exceeds a preset ratio threshold, merges an iterative training information set in the iterative training database with the initial training information set in the initial training database to generate an updated training information set;

And carrying out iterative training on the preset post description information generation model by utilizing the updated training information set, wherein each training step comprises the steps of e to h:

Step e: mapping each feature Word in the updated training information set to a representation in the continuous vector space by using the Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step f: inputting word embedding vectors corresponding to the feature words into the sequence encoder to generate hidden representations, and determining a hidden representation set corresponding to the updated training information set according to the hidden representations corresponding to the feature words;

step g: inputting the hidden representation set corresponding to the updated training information set to the sequence decoder to generate updated position description information;

step h: using the sequence cross entropy loss as a loss function, and determining a second loss difference value according to the post description information in the newly-added iteration training information and the updated generation post description information;

Updating model parameters of the preset post description information generation model according to the gradient of the second loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the second loss difference value is at a minimum value;

and carrying out model parameter adjustment on the preset post description information generation model according to the model parameters when the second loss difference value is at the minimum value so as to generate an updated preset post description information generation model.

In a second aspect, the application provides a recruitment information management platform, which is arranged on a cloud server and is in communication connection with each recruitment terminal in a recruitment terminal set;

the recruitment information management platform receives first post original information sent by a first recruitment terminal in the recruitment terminal set, wherein the first post original information comprises a first feature word set, and the first feature word set comprises a plurality of post feature words;

The recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to the first recruitment terminal, wherein the preset post description information generation model is a model generated based on training of a preset deep neural network model;

the recruitment information management platform receives first post feedback information sent by the first recruitment terminal in response to the first post description information;

If the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iteration training information according to the first post original information and the first post description information, and stores the first iteration training information as new iteration training information into an iteration training database corresponding to the preset deep neural network model, wherein the first iteration training information is configured with a first type of training feature labels.

Optionally, if the first post feedback information includes first post modification information and the first post release information, the recruitment information management platform releases the first post modification information, and generates second iterative training information according to the first post original information, the first post description information and the first post modification information, so as to store the second iterative training information as newly added iterative training information in the iterative training database, where the second iterative training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information.

Optionally, if the first post feedback information includes a first post modification request and the first post release information, the recruitment information management platform determines historical target post description information from a post release historical database according to the first feature word set, and determines a second recruitment terminal corresponding to the historical target post description information in the recruitment terminal set;

the recruitment information management platform sends first post correction information to the second recruitment terminal, wherein the first post correction information comprises the first post original information, the first post description information and the first post modification request;

The second recruitment terminal responds to the first post correction information and sends second post description information to the recruitment information management platform so that the recruitment information management platform sends the second post description information to the first recruitment terminal;

The first recruitment terminal responds to the second post description information and sends second post feedback information to the recruitment information management platform;

If the second post feedback information includes second post release information, the recruitment information management platform releases the second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information, so as to store the third iteration training information as new iteration training information into the iteration training database, wherein the third iteration training information is configured with the second class training feature tag, and the second post description information is post description information modified based on the first post description information.

Optionally, the recruitment information management platform utilizes a Word2Vec Word embedding model and based on the first feature Word setGenerating a first set of representations/>Wherein/>For the first feature word set/>/>Personal characteristic words,/>For the first feature word set/>/>Word embedding vector corresponding to each characteristic word,/>For the first feature word set/>The number of feature words in (a);

The recruitment information management platform utilizes jieba word segmentation tools and generates a first word segmentation set according to post release history information in the post release history database And determining the first Word segmentation set/>, by using Word2Vec Word embedding modelCorresponding second set of representations/>Wherein/>For the first word segment set/>/>Word segmentation,/>For the first word segment set/>/>Word embedding vector corresponding to each word segment,/>For the first word segment set/>The number of word segments in the database;

The recruitment information management platform calculates the first feature word set />Word embedding vector/>, corresponding to each feature wordAnd the second representation set/>Cosine similarity among the embedded vectors of each word;

If the word is embedded in the vector With word embedding vector/>Cosine similarity between/>If the recruitment information management platform is larger than the preset cosine similarity threshold, the recruitment information management platform determines the/>Individual feature words/>At the first word segmentation set/>The characteristic feature words exist in the Chinese character string;

The recruitment information management platform traverses the first set of representations Each word of the characteristic words is embedded into the vector, and if the quantity of the characteristic words is confirmed to exist/>, the characteristic words are representedAnd the first feature word set/>Number of feature words in/>The ratio of the post release historical information to the historical post description information to be selected is larger than a preset ratio threshold, and the historical post description information to be selected/>Wherein/>Candidate set/>, for the historical post description information/>Post release history information,/>Candidate set for the historical post description informationTotal number of post release history information;

The recruitment information management platform calculates the first feature word set according to formula (1) And/>Post release history information/>Cosine similarity mean value between/>The formula (1) is:

（1）；

wherein, To at/>Post release history information/>The corresponding second representation set/>Chinese and word embedding vector/>The word with the highest cosine similarity is embedded into the vector;

the recruitment information management platform determines a maximum cosine similarity average value according to a formula (2) The formula (2) is:

（2）；

The recruitment information management platform determines the maximum cosine similarity average value The corresponding post release history information is the history target post description information.

Optionally, the recruitment information management platform determines a call ratio corresponding to the published first post description information, wherein the call ratio is a ratio between the resume delivery number corresponding to the first post description information and the information browsing number;

and the recruitment information management platform determines that the calling ratio is smaller than a preset calling ratio threshold value.

Optionally, before the recruitment information management platform generates the first post description information according to the first post original information and the preset post description information generation model, the recruitment information management platform further includes:

creating an initial training database on the recruitment information management platform, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial characteristic word set and initial calibration post description information corresponding to the initial characteristic word set;

training the preset depth neural network model by using the initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder built on the basis of a cyclic neural network or a long-short-time memory network, and the sequence decoder is a decoder built on the basis of the cyclic neural network or an attention mechanism; wherein the step of each training comprises steps a to d:

step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

Step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of the preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at a minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate the preset post description information generation model.

Optionally, the recruitment information management platform determines the number of new information corresponding to the new iterative training information in the iterative training database, and if the number of new information exceeds a preset information number threshold, and a ratio between the number of new information of the second type corresponding to the new iterative training information configured with the second type training feature tag and the number of new information exceeds a preset ratio threshold, merges an iterative training information set in the iterative training database with the initial training information set in the initial training database to generate an updated training information set;

And carrying out iterative training on the preset post description information generation model by utilizing the updated training information set, wherein each training step comprises the steps of e to h:

Step e: mapping each feature Word in the updated training information set to a representation in the continuous vector space by using the Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step f: inputting word embedding vectors corresponding to the feature words into the sequence encoder to generate hidden representations, and determining a hidden representation set corresponding to the updated training information set according to the hidden representations corresponding to the feature words;

step g: inputting the hidden representation set corresponding to the updated training information set to the sequence decoder to generate updated position description information;

step h: using the sequence cross entropy loss as a loss function, and determining a second loss difference value according to the post description information in the newly-added iteration training information and the updated generation post description information;

Updating model parameters of the preset post description information generation model according to the gradient of the second loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the second loss difference value is at a minimum value;

and carrying out model parameter adjustment on the preset post description information generation model according to the model parameters when the second loss difference value is at the minimum value so as to generate an updated preset post description information generation model.

In a third aspect, the present application provides an electronic device comprising:

A processor; and

A memory for storing executable instructions of the processor;

Wherein the processor is configured to perform any one of the possible methods described in the first aspect via execution of the executable instructions.

In a fourth aspect, the present application provides a computer readable storage medium having stored therein computer executable instructions which when executed by a processor are adapted to carry out any one of the possible methods described in the first aspect.

According to the cloud service-based recruitment information intelligent processing method and platform, first post original information is sent to the recruitment information management platform through the first recruitment terminal, so that the recruitment information management platform generates first post description information according to the first post original information and preset post description information, the first post description information is sent to the first recruitment terminal, the first recruitment terminal responds to the first post description information, first post feedback information is sent to the recruitment information management platform, if the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iteration training information according to the first post original information and the first post description information, stores the first iteration training information as new iteration training information into an iteration training database corresponding to a preset depth neural network model, and then confirms release of the first post description information at the recruitment terminal, and then the first iteration training information is added into the iteration training database corresponding to the preset depth neural network model, so that the actual training information is updated according to the iteration training database, and the actual requirement is achieved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is a flow chart of an intelligent processing method for recruitment information based on cloud services according to an example embodiment of the present application;

fig. 2 is a flow chart of an intelligent processing method for recruitment information based on cloud services according to another example embodiment of the present application;

Fig. 3 is a schematic diagram of a recruitment information management system according to an example embodiment of the present application;

fig. 4 is a schematic structural view of an electronic device according to an exemplary embodiment of the present application.

Specific embodiments of the present application have been shown by way of the above drawings and will be described in more detail below. The drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but rather to illustrate the inventive concepts to those skilled in the art by reference to the specific embodiments.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.

Fig. 1 is a flow chart illustrating an intelligent processing method for recruitment information based on cloud services according to an example embodiment of the present application. As shown in fig. 1, the method provided in this embodiment includes:

s101, a first recruitment terminal in a recruitment terminal set sends first post original information to a recruitment information management platform.

The intelligent processing method of recruitment information based on cloud service provided by the embodiment can be applied to a recruitment information management system, wherein the recruitment information management system comprises a recruitment terminal set and an information management platform arranged on a cloud server, and the information management platform is in communication connection with each recruitment terminal in the recruitment terminal set.

In this step, a first recruitment terminal in the recruitment terminal set sends first post original information to the recruitment information management platform, where the first post original information includes a first feature word set, and the first feature word set includes a plurality of post feature words, where the post feature words may be, for example, software, an engineer, JAVA, a server, background development, and the like.

S102, the recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to a first recruitment terminal.

In this step, the recruitment information management platform may generate a model according to the first post original information and the preset post description information to generate first post description information, and send the first post description information to the first recruitment terminal, where the preset post description information generation model is a model generated based on training of a preset deep neural network model. The pre-set post description information generation model may be specifically a model generated based on convolutional neural network (Convolutional Neural Network, CNN), cyclic neural network (Recurrent Neural Network, RNN), long Short-Term Memory (LSTM), and bidirectional cyclic neural network (Bidirectional Recurrent Neural Network, biRNN) training.

And S103, the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform.

In this step, the first recruitment terminal may send first post feedback information to the recruitment information management platform in response to the first post description information.

And S104, if the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information and generates first iteration training information according to the first post original information and the first post description information.

Specifically, if the first post feedback information includes first post release information, the recruitment information management platform releases the first post description information, and generates first iterative training information according to the first post original information and the first post description information, so as to store the first iterative training information as newly added iterative training information into an iterative training database corresponding to a preset deep neural network model, wherein the first iterative training information is configured with a first type training feature tag.

In this embodiment, first post original information is sent to a recruitment information management platform through a first recruitment terminal, so that the recruitment information management platform generates a model according to the first post original information and preset post description information to generate first post description information, and sends the first post description information to the first recruitment terminal, so that the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform, if the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iterative training information according to the first post original information and the first post description information, stores the first iterative training information as new iterative training information in an iterative training database corresponding to a preset deep neural network model, automatically generates post description information through post feature words, and adds the generated post description information to the iterative training database after the recruitment terminal confirms release, so that the iterative training information can be more and more matched with the actual training model, and further achieves the purposes of updating the iterative training model.

Fig. 2 is a flow chart of an intelligent processing method for recruitment information based on cloud services according to another example embodiment of the present application. As shown in fig. 2, the method provided in this embodiment includes:

S201, a first recruitment terminal in the recruitment terminal set sends first post original information to a recruitment information management platform.

In this step, a first recruitment terminal in the recruitment terminal set sends first post original information to the recruitment information management platform, where the first post original information includes a first feature word set, and the first feature word set includes a plurality of post feature words, where the post feature words may be, for example, software, an engineer, JAVA, a server, background development, and the like.

S202, the recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to a first recruitment terminal.

In this step, the recruitment information management platform may generate a model according to the first post original information and the preset post description information to generate first post description information, and send the first post description information to the first recruitment terminal, where the preset post description information generation model is a model generated based on training of a preset deep neural network model.

Optionally, before the recruitment information management platform generates the first post description information according to the first post original information and the preset post description information generation model, the method further includes:

Creating an initial training database on a recruitment information management platform, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial feature word set and initial calibration post description information corresponding to the initial feature word set;

Training a preset depth neural network model by using an initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder established based on a cyclic neural network or a long-short-term memory network, and the sequence decoder is a decoder established based on the cyclic neural network or a attention mechanism; wherein the step of each training comprises steps a to d:

Step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of a preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at the minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate a preset post description information generation model.

And S203, the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform.

In this step, the first recruitment terminal may send first post feedback information to the recruitment information management platform in response to the first post description information.

S204, if the first post feedback information comprises first post modification information and first post release information, the recruitment information management platform releases the first post modification information, and generates second iteration training information according to the first post original information, the first post description information and the first post modification information.

Specifically, if the first post feedback information includes first post modification information and first post release information, the recruitment information management platform releases the first post modification information, and generates second iteration training information according to the first post original information, the first post description information and the first post modification information, so as to store the second iteration training information as new iteration training information into the iteration training database, wherein the second iteration training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information.

S205, if the first post feedback information comprises a first post modification request and first post release information, the recruitment information management platform determines historical target post description information from a post release historical database according to the first feature word set, and determines a second recruitment terminal corresponding to the historical target post description information in the recruitment terminal set.

Specifically, the recruitment information management platform utilizes a Word2Vec Word embedding model and based on the first feature Word setGenerating a first set of representations/>Wherein/>For the first feature word set/>/>Personal characteristic words,/>For the first feature word set/>/>Word embedding vector corresponding to each characteristic word,/>For the first feature word set/>The number of feature words in (a);

The recruitment information management platform utilizes jieba word segmentation tools and generates a first word segmentation set according to post release history information in the post release history database And determining the first Word segmentation set/>, by using Word2Vec Word embedding modelCorresponding second set of representations/>Wherein/>For the first word segment set/>/>Word segmentation,/>For the first word segment set/>/>Word embedding vector corresponding to each word segment,/>For the first word segment set/>The number of word segments in the database;

The recruitment information management platform calculates the first feature word set />Word embedding vector/>, corresponding to each feature wordAnd the second representation set/>Cosine similarity among the embedded vectors of each word;

If words are embedded in vectors With word embedding vector/>Cosine similarity between/>If the recruitment information management platform is larger than the preset cosine similarity threshold, the recruitment information management platform determines the/>Individual feature words/>At the first word segmentation set/>The characteristic feature words exist in the Chinese character string;

traversing the first representation set by the recruitment information management platform The recruitment information management platform traverses each word embedding vector in the first representation set, and if the number of characterization feature words/>, which are determined to exist, is determinedAnd the first characteristic word setNumber of feature words in/>The ratio of the feature words to the feature words is larger than a preset ratio threshold, the post sending recruitment information management platform traverses each word embedding vector in the first representation set, and if the ratio of the number of the determined feature words to the number of the feature words in the first feature word set is larger than the preset ratio threshold, the post release historical information is added to a historical post description information candidate set, wherein the historical post description information candidate set is selected by the post sending recruitment information management platformWherein/>Candidate set/>, for the historical post description information/>Post release history information,/>Candidate set/>, for the historical post description informationTotal number of post release history information;

the recruitment information management platform calculates the first feature word set according to the formula (1) And/>Post release history information/>Cosine similarity mean value between/>The formula (1) is:

（1）；

wherein, To at/>Post release history information/>The corresponding second representation set/>Chinese and word embedding vector/>The word with the highest cosine similarity is embedded into the vector;

The recruitment information management platform determines the average value of the maximum cosine similarity according to the formula (2) The formula (2) is:

（2）；

The recruitment information management platform determines the average value of the maximum cosine similarity The corresponding post release history information is the history target post description information.

S206, the second recruitment terminal responds to the first post correction information and sends second post description information to the recruitment information management platform.

Specifically, the second recruitment terminal responds to the first position correction information and sends second position description information to the recruitment information management platform, so that the recruitment information management platform sends the second position description information to the first recruitment terminal.

In addition, before the recruitment information management platform sends the first post correction information to the second recruitment terminal, the method may further include:

The recruitment information management platform determines a recall ratio corresponding to the published first post description information, wherein the recall ratio is a ratio between the resume delivery quantity corresponding to the first post description information and the information browsing quantity; and the recruitment information management platform determines that the recall ratio is less than a preset recall ratio threshold. When the recruitment information management platform determines that the recruitment ratio is smaller than a preset recruitment ratio threshold, namely the situation that the ratio between the resume delivery quantity and the information browsing quantity is smaller occurs, the situation that the published first post description information is deviated from the actual situation or is inaccurately described is indicated, and at the moment, other recruitment terminals can be triggered to correct the post description information, so that the accuracy of recruitment information publishing and the recruitment efficiency are improved.

S207, the first recruitment terminal responds to the second post description information and sends second post feedback information to the recruitment information management platform.

S208, if the second post feedback information comprises second post release information, the recruitment information management platform releases the second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information.

If the second post feedback information comprises second post release information, the recruitment information management platform releases second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information, so as to store the third iteration training information into the iteration training database as newly added iteration training information, wherein the third iteration training information is configured with a second type training feature tag, and the second post description information is post description information modified based on the first post description information.

In addition, after the first iterative training information is stored in the iterative training database corresponding to the preset deep neural network model, the recruitment information management platform determines the number of the newly-added information corresponding to the newly-added iterative training information in the iterative training database, and if the number of the newly-added information exceeds a preset information number threshold and the ratio between the number of the newly-added information and the number of the newly-added information corresponding to the newly-added iterative training information configured with the second-class training feature tag exceeds a preset ratio threshold, the iterative training information set in the iterative training database is combined with the initial training information set in the initial training database to generate an updated training information set. It is worth to describe that the accuracy of the information generated by the model is further improved by setting up that the ratio between the number of the second type of added information and the number of the new added information exceeds a preset ratio threshold as a training triggering condition, combining the iterative training information set in the iterative training database with the initial training information set in the initial training database, and then training the model.

Specifically, the training information set is updated to perform iterative training on the preset post description information generation model, where each training step includes steps e to h:

Step e: mapping each feature Word in the updated training information set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step f: the word embedding vectors corresponding to the feature words are input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the training information set is determined to be updated according to the hidden representations corresponding to the feature words;

Step g: inputting the hidden representation set corresponding to the updated training information set to a sequence decoder to generate updated generation post description information;

step h: using the sequence cross entropy loss as a loss function, and determining a second loss difference value according to post description information in the newly-added iteration training information and the updated generated post description information;

updating model parameters of a preset post description information generation model according to the gradient of the second loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the second loss difference value is at the minimum value;

And carrying out model parameter adjustment on the preset post description information generation model according to the model parameters when the second loss difference value is at the minimum value so as to generate an updated preset post description information generation model.

Fig. 3 is a schematic diagram of a recruitment information management system according to an example embodiment of the present application. As shown in fig. 3, the recruitment information management system 300 provided in this embodiment includes: the recruitment terminal comprises a recruitment terminal set 310 and a recruitment information management platform 320 arranged on a cloud server, wherein the recruitment information management platform 320 is in communication connection with each recruitment terminal in the recruitment terminal set 310; the method comprises the following steps:

A first recruitment terminal in the recruitment terminal set sends first post original information to the recruitment information management platform 320, wherein the first post original information comprises a first feature word set, and the first feature word set comprises a plurality of post feature words;

The recruitment information management platform 320 generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to the first recruitment terminal, wherein the preset post description information generation model is a model generated based on training of a preset deep neural network model;

The first recruitment terminal transmits first post feedback information to the recruitment information management platform 320 in response to the first post description information;

If the first post feedback information includes first post release information, the recruitment information management platform 320 releases the first post description information, and generates first iterative training information according to the first post original information and the first post description information, so as to store the first iterative training information as new iterative training information in an iterative training database corresponding to the preset deep neural network model, where the first iterative training information is configured with a first type of training feature tag.

Optionally, if the first post feedback information includes first post modification information and the first post release information, the recruitment information management platform 320 releases the first post modification information, and generates second iterative training information according to the first post original information, the first post description information and the first post modification information, so as to store the second iterative training information as newly added iterative training information in the iterative training database, where the second iterative training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information.

Optionally, if the first post feedback information includes a first post modification request and the first post release information, the recruitment information management platform 320 determines, according to the first feature word set, historical target post description information from a post release historical database, and determines a second recruitment terminal corresponding to the historical target post description information in the recruitment terminal set;

the recruitment information management platform 320 sends first post modification information to the second recruitment terminal, wherein the first post modification information includes the first post original information, the first post description information, and the first post modification request;

The second recruitment terminal transmits second post description information to the recruitment information management platform 320 in response to the first post correction information, so that the recruitment information management platform 320 transmits the second post description information to the first recruitment terminal;

the first recruitment terminal transmits second post feedback information to the recruitment information management platform 320 in response to the second post description information;

If the second post feedback information includes second post release information, the recruitment information management platform 320 releases the second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information, so as to store the third iteration training information as new iteration training information into the iteration training database, wherein the third iteration training information is configured with the second class training feature tag, and the second post description information is post description information modified based on the first post description information.

Optionally, the recruitment information management platform 320 utilizes a Word2Vec Word embedding model and is based on the first set of feature wordsGenerating a first set of representations/>Wherein/>For the first feature word set/>/>Personal characteristic words,/>For the first feature word set/>/>Word embedding vector corresponding to each characteristic word,/>For the first feature word set/>The number of feature words in (a);

The recruitment information management platform 320 utilizes jieba a word segmentation tool and generates a first word segmentation set based on post publication history information in the post publication history database And determining the first Word segmentation set/>, by using Word2Vec Word embedding modelCorresponding second representation setWherein/>For the first word segment set/>/>Word segmentation,/>For the first word segment set/>/>Word embedding vector corresponding to each word segment,/>For the first word segment set/>The number of word segments in the database;

The recruitment information management platform 320 calculates the first set of feature words />Word embedding vector/>, corresponding to each feature wordAnd the second representation set/>Cosine similarity among the embedded vectors of each word;

If the word is embedded in the vector With word embedding vector/>Cosine similarity between/>If the recruitment information management platform 320 is greater than the preset cosine similarity threshold, the recruitment information management platform determines the/>Individual feature words/>At the first word segmentation set/>The characteristic feature words exist in the Chinese character string;

The recruitment information management platform 320 traverses the first set of representations Each word of the characteristic words is embedded into the vector, and if the quantity of the characteristic words is confirmed to exist/>, the characteristic words are representedAnd the first feature word set/>Number of feature words in/>The ratio of the post release historical information to the historical post description information to be selected is larger than a preset ratio threshold, and the historical post description information to be selected/>Wherein/>Candidate set/>, for the historical post description information/>Post release history information,/>Candidate set/>, for the historical post description informationTotal number of post release history information;

the recruitment information management platform 320 calculates the first set of feature words according to equation (1) And/>Post release history information/>Cosine similarity mean value between/>The formula (1) is:

（1）；

wherein, To at/>Post release history information/>The corresponding second representation set/>Chinese and word embedding vector/>The word with the highest cosine similarity is embedded into the vector;

the recruitment information management platform 320 determines a maximum cosine similarity average according to equation (2) The formula (2) is:

（2）；/>

the recruitment information management platform 320 determines the maximum cosine similarity average The corresponding post release history information is the history target post description information.

Optionally, the recruitment information management platform 320 determines a call ratio corresponding to the first post description information after being published, where the call ratio is a ratio between a resume delivery number corresponding to the first post description information and an information browsing number;

The recruitment information management platform 320 determines that the calling ratio is less than a preset calling ratio threshold.

Optionally, before the recruitment information management platform 320 generates the first post description information according to the first post original information and the preset post description information generation model, the method further includes:

creating an initial training database on the recruitment information management platform 320, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial feature word set and initial calibration post description information corresponding to the initial feature word set;

training the preset depth neural network model by using the initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder built on the basis of a cyclic neural network or a long-short-time memory network, and the sequence decoder is a decoder built on the basis of the cyclic neural network or an attention mechanism; wherein the step of each training comprises steps a to d:

step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

Step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of the preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at a minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate the preset post description information generation model.

Optionally, the recruitment information management platform 320 determines the number of new information corresponding to the new iterative training information in the iterative training database, and if the number of new information exceeds a preset information number threshold, and a ratio between the number of new information of the second type corresponding to the new iterative training information configured with the second type training feature tag and the number of new information exceeds a preset ratio threshold, merges the iterative training information set in the iterative training database with the initial training information set in the initial training database to generate an updated training information set;

And carrying out iterative training on the preset post description information generation model by utilizing the updated training information set, wherein each training step comprises the steps of e to h:

Step e: mapping each feature Word in the updated training information set to a representation in the continuous vector space by using the Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step f: inputting word embedding vectors corresponding to the feature words into the sequence encoder to generate hidden representations, and determining a hidden representation set corresponding to the updated training information set according to the hidden representations corresponding to the feature words;

step g: inputting the hidden representation set corresponding to the updated training information set to the sequence decoder to generate updated position description information;

step h: using the sequence cross entropy loss as a loss function, and determining a second loss difference value according to the post description information in the newly-added iteration training information and the updated generation post description information;

Updating model parameters of the preset post description information generation model according to the gradient of the second loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the second loss difference value is at a minimum value;

and carrying out model parameter adjustment on the preset post description information generation model according to the model parameters when the second loss difference value is at the minimum value so as to generate an updated preset post description information generation model.

Fig. 4 is a schematic structural view of an electronic device according to an exemplary embodiment of the present application. As shown in fig. 4, an electronic device 400 provided in this embodiment includes: a processor 401 and a memory 402; wherein:

a memory 402 for storing a computer program, which memory may also be a flash memory.

A processor 401 for executing the execution instructions stored in the memory to implement the steps in the above method. Reference may be made in particular to the description of the embodiments of the method described above.

Alternatively, the memory 402 may be separate or integrated with the processor 401.

When the memory 402 is a device separate from the processor 401, the electronic apparatus 400 may further include:

A bus 403 for connecting the memory 402 and the processor 401.

The present embodiment also provides a readable storage medium having a computer program stored therein, which when executed by at least one processor of an electronic device, performs the methods provided by the various embodiments described above.

The present embodiment also provides a program product comprising a computer program stored in a readable storage medium. The computer program may be read from a readable storage medium by at least one processor of an electronic device, and executed by the at least one processor, causes the electronic device to implement the methods provided by the various embodiments described above.

Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

It is to be understood that the application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (8)

1. The intelligent recruitment information processing method based on the cloud service is characterized by being applied to a recruitment information management system, wherein the recruitment information management system comprises a recruitment terminal set and a recruitment information management platform arranged on a cloud server, and the recruitment information management platform is in communication connection with each recruitment terminal in the recruitment terminal set; the method comprises the following steps:

A first recruitment terminal in the recruitment terminal set sends first post original information to the recruitment information management platform, wherein the first post original information comprises a first feature word set, and the first feature word set comprises a plurality of post feature words;

The recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to the first recruitment terminal, wherein the preset post description information generation model is a model generated based on training of a preset deep neural network model;

the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform;

If the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iteration training information according to the first post original information and the first post description information, and stores the first iteration training information as newly added iteration training information into an iteration training database corresponding to the preset deep neural network model, wherein the first iteration training information is configured with a first type of training feature labels;

After the first recruitment terminal responds to the first post description information and sends first post feedback information to the recruitment information management platform, the method further comprises:

If the first post feedback information comprises first post modification information and the first post release information, the recruitment information management platform releases the first post modification information, and generates second iteration training information according to the first post original information, the first post description information and the first post modification information so as to store the second iteration training information as newly added iteration training information into the iteration training database, wherein the second iteration training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information;

before the recruitment information management platform generates the first post description information according to the first post original information and the preset post description information generation model, the recruitment information management platform further comprises:

creating an initial training database on the recruitment information management platform, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial characteristic word set and initial calibration post description information corresponding to the initial characteristic word set;

training the preset depth neural network model by using the initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder built on the basis of a cyclic neural network or a long-short-time memory network, and the sequence decoder is a decoder built on the basis of the cyclic neural network or an attention mechanism; wherein the step of each training comprises steps a to d:

step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

Step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of the preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at a minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate the preset post description information generation model.

2. The intelligent processing method of recruitment information based on cloud service according to claim 1, further comprising, after the first recruitment terminal sends first post feedback information to the recruitment information management platform in response to the first post description information:

If the first post feedback information comprises a first post modification request and the first post release information, the recruitment information management platform determines historical target post description information from a post release historical database according to the first feature word set, and determines a second recruitment terminal corresponding to the historical target post description information in the recruitment terminal set;

the recruitment information management platform sends first post correction information to the second recruitment terminal, wherein the first post correction information comprises the first post original information, the first post description information and the first post modification request;

The second recruitment terminal responds to the first post correction information and sends second post description information to the recruitment information management platform so that the recruitment information management platform sends the second post description information to the first recruitment terminal;

The first recruitment terminal responds to the second post description information and sends second post feedback information to the recruitment information management platform;

If the second post feedback information includes second post release information, the recruitment information management platform releases the second post description information, and generates third iteration training information according to the first post original information, the first post description information and the second post description information, so as to store the third iteration training information as new iteration training information into the iteration training database, wherein the third iteration training information is configured with the second class training feature tag, and the second post description information is post description information modified based on the first post description information.

3. The intelligent processing method of recruitment information based on cloud service of claim 2, wherein the recruitment information management platform determines historical target post description information from a post publication history database according to the first feature word set, comprising:

The recruitment information management platform utilizes a Word2Vec Word embedding model and based on the first feature Word set Generating a first set of representations/>Wherein/>For the first feature word set/>/>Personal characteristic words,/>For the first feature word set/>/>Word embedding vector corresponding to each characteristic word,/>For the first feature word set/>The number of feature words in (a);

The recruitment information management platform utilizes jieba word segmentation tools and generates a first word segmentation set according to post release history information in the post release history database And determining the first Word segmentation set/>, by using Word2Vec Word embedding modelCorresponding second representation setWherein/>For the first word segment set/>/>Word segmentation,/>For the first word segment set/>/>Word embedding vector corresponding to each word segment,/>For the first word segment set/>The number of word segments in the database;

The recruitment information management platform calculates the first feature word set />Word embedding vector/>, corresponding to each feature wordAnd the second representation set/>Cosine similarity among the embedded vectors of each word;

If the word is embedded in the vector With word embedding vector/>Cosine similarity between/>If the recruitment information management platform is larger than the preset cosine similarity threshold, the recruitment information management platform determines the/>Individual feature words/>At the first word segmentation set/>The characteristic feature words exist in the Chinese character string;

The recruitment information management platform traverses the first set of representations Each word of the characteristic words is embedded into the vector, and if the quantity of the characteristic words is confirmed to exist/>, the characteristic words are representedAnd the first feature word set/>Number of feature words in/>The ratio of the post release historical information to the historical post description information to be selected is larger than a preset ratio threshold, and the historical post description information to be selected/>Wherein/>Candidate set/>, for the historical post description information/>Post release history information,/>Candidate set/>, for the historical post description informationTotal number of post release history information;

The recruitment information management platform calculates the first feature word set according to formula (1) And/>Post release history information/>Cosine similarity mean value between/>The formula (1) is:

（1）;

wherein, To at/>Post release history information/>The corresponding second representation set/>Chinese and word embedding vector/>The word with the highest cosine similarity is embedded into the vector;

the recruitment information management platform determines a maximum cosine similarity average value according to a formula (2) The formula (2) is:

（2）；

The recruitment information management platform determines the maximum cosine similarity average value The corresponding post release history information is the history target post description information.

4. The intelligent processing method of recruitment information based on cloud service according to claim 2 or 3, further comprising, before the recruitment information management platform sends the first post correction information to the second recruitment terminal:

The recruitment information management platform determines the corresponding calling ratio of the published first post description information, wherein the calling ratio is the ratio between the delivery number of resume corresponding to the first post description information and the information browsing number;

and the recruitment information management platform determines that the calling ratio is smaller than a preset calling ratio threshold value.

5. The intelligent processing method of recruitment information based on cloud service according to claim 1, further comprising, after storing the first iterative training information in an iterative training database corresponding to the preset deep neural network model:

The recruitment information management platform determines the number of newly-added information corresponding to newly-added iterative training information in the iterative training database, and if the number of newly-added information exceeds a preset information number threshold, and the ratio between the number of newly-added information of the second type corresponding to newly-added iterative training information configured with the second type training feature tag and the number of newly-added information exceeds a preset ratio threshold, merges an iterative training information set in the iterative training database with the initial training information set in the initial training database to generate an updated training information set;

And carrying out iterative training on the preset post description information generation model by utilizing the updated training information set, wherein each training step comprises the steps of e to h:

Step e: mapping each feature Word in the updated training information set to a representation in the continuous vector space by using the Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step f: inputting word embedding vectors corresponding to the feature words into the sequence encoder to generate hidden representations, and determining a hidden representation set corresponding to the updated training information set according to the hidden representations corresponding to the feature words;

step g: inputting the hidden representation set corresponding to the updated training information set to the sequence decoder to generate updated position description information;

step h: using the sequence cross entropy loss as a loss function, and determining a second loss difference value according to the post description information in the newly-added iteration training information and the updated generation post description information;

Updating model parameters of the preset post description information generation model according to the gradient of the second loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the second loss difference value is at a minimum value;

and carrying out model parameter adjustment on the preset post description information generation model according to the model parameters when the second loss difference value is at the minimum value so as to generate an updated preset post description information generation model.

6. The recruitment information management platform is characterized by being arranged on a cloud server and in communication connection with each recruitment terminal in a recruitment terminal set;

the recruitment information management platform receives first post original information sent by a first recruitment terminal in the recruitment terminal set, wherein the first post original information comprises a first feature word set, and the first feature word set comprises a plurality of post feature words;

The recruitment information management platform generates first post description information according to the first post original information and a preset post description information generation model, and sends the first post description information to the first recruitment terminal, wherein the preset post description information generation model is a model generated based on training of a preset deep neural network model;

the recruitment information management platform receives first post feedback information sent by the first recruitment terminal in response to the first post description information;

If the first post feedback information comprises first post release information, the recruitment information management platform releases the first post description information, generates first iteration training information according to the first post original information and the first post description information, and stores the first iteration training information as newly added iteration training information into an iteration training database corresponding to the preset deep neural network model, wherein the first iteration training information is configured with a first type of training feature labels;

If the first post feedback information comprises first post modification information and the first post release information, the recruitment information management platform releases the first post modification information, and generates second iteration training information according to the first post original information, the first post description information and the first post modification information so as to store the second iteration training information as newly added iteration training information into the iteration training database, wherein the second iteration training information is configured with a second type training feature tag, and the first post modification information is post description information modified based on the first post description information;

creating an initial training database on the recruitment information management platform, wherein the initial training database comprises an initial training information set, and each piece of initial training information in the initial training information set comprises an initial characteristic word set and initial calibration post description information corresponding to the initial characteristic word set;

training the preset depth neural network model by using the initial training information set, wherein the preset depth neural network model comprises a sequence encoder and a sequence decoder, the sequence encoder is connected with the sequence decoder end to end, the sequence encoder is an encoder built on the basis of a cyclic neural network or a long-short-time memory network, and the sequence decoder is a decoder built on the basis of the cyclic neural network or an attention mechanism; wherein the step of each training comprises steps a to d:

step a: mapping each initial feature Word in the initial feature Word set to a representation in a continuous vector space by using a Word2Vec Word embedding model, wherein the representation in the continuous vector space is used for determining a corresponding Word embedding vector;

Step b: the word embedding vector corresponding to each initial feature word is input to a sequence encoder to generate hidden representations, and a hidden representation set corresponding to the initial feature word set is determined according to the hidden representations corresponding to each initial feature word;

Step c: inputting the hidden representation set to a sequence decoder to generate initial generation post description information;

Step d: using the sequence cross entropy loss as a loss function, and determining a first loss difference value according to the initial calibration post description information and the initial generation post description information;

Updating model parameters of the preset deep neural network model according to the gradient of the first loss difference value in each training by using a back propagation algorithm, and determining the model parameters when the first loss difference value is at a minimum value;

And carrying out model parameter adjustment on the preset deep neural network model according to the model parameters when the first loss difference value is at the minimum value so as to generate the preset post description information generation model.

7. An electronic device, comprising:

A processor; and

A memory for storing executable instructions of the processor;

Wherein the processor is configured to perform the method of any one of claims 1 to 5 via execution of the executable instructions.

8. A computer readable storage medium having stored therein computer executable instructions which when executed by a processor are adapted to carry out the method of any one of claims 1 to 5.