CN115795125A - Searching method, device, equipment and medium applied to project management software - Google Patents
Searching method, device, equipment and medium applied to project management software Download PDFInfo
- Publication number
- CN115795125A CN115795125A CN202310060044.2A CN202310060044A CN115795125A CN 115795125 A CN115795125 A CN 115795125A CN 202310060044 A CN202310060044 A CN 202310060044A CN 115795125 A CN115795125 A CN 115795125A
- Authority
- CN
- China
- Prior art keywords
- node
- search
- weight
- nodes
- operation node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a searching method, a searching device, searching equipment and searching media applied to project management software. The searching method comprises the following steps: accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes; the input node generates a search condition according to the query condition input to the project management software and sends the search condition to the operation node; the operation node performs operation according to the search condition, generates a data set and transmits the data set to the next operation node and the output node, wherein the data set comprises the weight of the operation node and the search result; the next operation node repeatedly executes the same operation until the operation is not needed to be continued; and the output node sorts the search results according to the weight of the operation node and sequentially outputs the search results to the terminal. The method can quickly screen out the target result, and does not need to specially develop and train for the specified search scene, thereby saving the development and maintenance cost.
Description
Technical Field
The invention belongs to the field of intelligent search, and particularly relates to a search method which is applied to project management software and has various changing conditions in a use environment.
Background
The project management software is used for development, operation, monitoring, maintenance and the like of software. Most project management software has a search function for searching contents designated by a user. However, the search logic applied may be different for each designated content. For example, when searching for a certain software using project management class software, if there are multiple versions of the same software, the search logic focuses on the update date of the software; when searching for log information using the item management-type software, the search logic focuses on the importance level of the log information. Thus, when there are a plurality of results satisfying the condition, the display order of the plurality of results greatly affects the experience of the user. Therefore, in the existing Search function, a learning model based on Neural Architecture Search (NAS) is basically added to analyze results that a user wants to see best and to preferentially display the results.
However, the learning model based on neural structure search has the following problems:
1. the learning model based on neural structure search has a slow growth rate, and needs to complete training through a large amount of data, and if the number of times of using the search function by a user is small, the learning model cannot obtain enough data to learn and grow.
2. Since search scenarios are varied (e.g., searching software information, searching log data, searching values satisfying search conditions in real-time interactive data, etc.), a set of related learning models needs to be written separately for each scenario, which increases development and maintenance costs of project management software.
3. Due to the fact that the learning model based on the neural structure search is slow in growth speed and different in search habits of each user, when the user changes, the learning model needs to spend a long time for retraining, and search results cannot be obtained in an express mode.
Disclosure of Invention
Aiming at the problems, the invention provides a searching method applied to project management software, which is used for improving the traditional learning model based on neural structure search.
In one aspect of the present invention, a search method applied to project management software is provided, including:
accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently;
when a query condition is input into the project management software, the input node generates a search condition according to the query condition and sends the search condition to the operation node;
the operation node operates according to the search condition, generates a data set and transmits the data set to a next operation node and an output node, wherein the data set comprises a weight of the operation node and a search result;
the next operation node repeatedly executes the same operation as the operation node until the learning model judges that the operation does not need to be continued; and
and the output node sorts the received search results in the data set according to the weight of the operation node, and sequentially outputs the search results to the terminal for display.
Optionally, the search method of the present invention further includes: and after the query is finished, emptying temporary data in the learning model.
Optionally, in the search method of the present invention, the data set of the operation node includes: the search condition of the operation node, the search result of the operation node, the path of the operation node and the weight of the operation node.
Optionally, in the search method of the present invention, the weight values included in the data set of the operation node are the weight values calculated latest after the operation of the operation node is performed, and a calculation formula is as follows:
the new weight value = (the weight value of the operation node ^ 2)/the weight value in the data set of the precursor node + the weight value in the data set of the precursor node.
Optionally, in the searching method of the present invention, when the user selects the output result with a low weight for multiple times, the step of recalculating the weight of the operation node and correcting the predecessor node and/or successor node of the operation node includes:
calculating the weights of all operation nodes on the path of the selected output result, marking the operation nodes as the model soldier nodes, and calculating the weights of the model soldier nodes by the following formula:
calculating a node weight = a current weight X, which is an important value of the searching condition;
calculating the weights of the operation nodes contained in other output results of which the weights are larger than the selected output result, and marking the operation nodes as non-standard soldier nodes, wherein the weight calculation formula of the non-standard soldier nodes is as follows:
operation node weight = current weight X the important value X12.73% of this search condition X the number of times this operation node uses in this output result;
when the data set corresponding to the selected output result is not the first output result in the plurality of output results displayed by the terminal, the weight of the model node is calculated again, and the weight calculation formula of the model node is calculated again as follows:
the operation node weight = the current weight X ((order number X1.375%) +100% when the data set of the operation node is output).
And taking the node of the operation node with the weight value larger than the minimum weight value in the model nodes in the non-model nodes as the successor node of the operation node with the minimum weight value.
Optionally, the searching method of the present invention is used in the case of maintaining software and searching infrequently.
In another aspect of the present invention, there is provided a search apparatus applied to item management software, including:
the access module is used for accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently;
the first generation module is used for enabling the input node to generate a search condition according to the query condition when the query condition is input into the project management software and sending the search condition to the operation node;
the second generation module is used for enabling the operation node to operate according to the search condition, generating a data set and transmitting the data set to a next operation node and an output node, wherein the data set comprises a weight of the operation node and a search result;
the execution module is used for enabling the next operation node to repeatedly execute the same operation as the operation node until the learning model judges that the operation does not need to be continued; and
and the output module is used for enabling the output node to sort the search results in the received data set according to the weight of the operation node and sequentially output the search results to the terminal for displaying.
In a further aspect of the present invention, there is provided a computer device comprising a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the searching method according to any of the above embodiments of the present invention.
In a further aspect of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, and the computer program is loaded and executed by a processor to implement the searching method according to any of the above embodiments of the present invention.
In the searching method, device, equipment and medium applied to the project management software, each operation node in the operation layer of the learning model can work independently, namely, as long as the input from the predecessor node exists, the operation can be carried out and the result is output without waiting for the input of all predecessor nodes, so that the output result of the operation node is increased. Therefore, the target result can be quickly screened out. In addition, because the operation nodes do not have strong dependence relationship with each other, the learning model has wider application scene than the traditional model, the use environment can have various changing conditions, the operation nodes can be multiplexed, special development and training for the specified scene are not needed, and the development cost is saved.
In addition, the weight of the operation node in the learning model and the predecessor node and successor node of the operation node can be continuously corrected along with the search operation, so that the weight of the operation node can be more accurately and intelligently calculated. When the user changes, the invention can also quickly adjust the weight of the operation node, thereby more quickly calculating the result desired by the user.
Thus, the advantages of the present invention include at least:
1. the learning model does not need to be developed again aiming at different search scenes, so that the development cost and the maintenance cost of project management software are reduced.
2. The learning model of the invention has fast growth speed, can complete the training of the learning model by a small amount of data, and does not need to train the model by a large amount of data in advance, therefore, even if the times of using the search function is few, the user can be provided with the expected search result.
3. Because the learning model of the invention can complete training by a small amount of data, when a user changes, the learning model can quickly calculate the search result according to the habit of the user.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
Fig. 1 is a flowchart of one embodiment of a search method applied to project management software according to the present invention.
Fig. 2 is a partial deployment diagram of application project management software according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of a node architecture of a learning model according to an embodiment of the present invention.
Fig. 4 is a flowchart of another embodiment of the search method applied to the project management software according to the present invention.
FIG. 5 is a schematic diagram of a workflow process for searching log information according to an embodiment of the invention.
FIG. 6 is a schematic structural diagram of a search apparatus applied to project management software according to an embodiment of the present invention.
Detailed Description
In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Those skilled in the art will appreciate that the following description of exemplary embodiments is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses. Moreover, the descriptions of the embodiments of the present invention emphasize the differences between the embodiments, and the same or similar parts may be referred to each other, so that the descriptions thereof are omitted for brevity.
It should also be understood that in embodiments of the present invention, "a plurality" may refer to two or more. The term "and/or" in the present invention is only an association relationship describing an associated object, and means that there may be three relationships, for example, a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone.
For ease of understanding, the following description will discuss relevant terms and concepts to which the present invention may be directed.
Neural Architecture Search (NAS for short) refers to a set of candidate Neural network nodes called a Search space is given, a subnetwork structure is searched from the set according to a Search strategy, performance evaluation strategy evaluation performance is used, and an evaluation result is returned to the Search strategy to adjust the next Neural structure selection and iterate until a Neural network meeting requirements is searched.
In the search function of the project management software, a NAS-based learning model is usually added to analyze the results that users most want to see and preferentially display the results. In some examples, operations implemented by the NAS may be abstracted into a directed acyclic graph formed without isolated nodes, the nodes representing local operations, the directed connections between the nodes representing flows of data, the nodes forming a layer of the learning model.
When the learning model based on the NAS is used for searching, if one operation layer has a plurality of predecessor nodes, the operation nodes on the operation layer need to wait for the outputs of all the predecessor nodes to reach the operation layer before performing operation, in other words, each operation node needs to wait for all the inputs of the operation node to be determined before starting the next operation, which greatly increases the time and the growth speed of the whole learning model for outputting the result once.
The invention reforms the traditional model based on NAS, and the basic principle of the realization is as follows: and (4) disassembling the operation nodes on the operation layer, so that each operation node of the operation layer can work in combination with other operation nodes or work independently. In other words, in the conventional model of NAS, the operation nodes in the operation layer need to be configured more closely and have stronger dependency relationship, but the improved model of the present invention reduces the strong dependency relationship between the operation nodes, so that each operation node can be combined with other operation nodes to work, or can work alone, for example, the operation can be performed according to some conditions, and it is not necessary to wait until all conditions are determined.
Moreover, the improved model of the invention can also correct the predecessor node and/or successor node of the operation node and/or modify the weight of the operation node according to the selection of the user on the search result.
Because each operation node of the operation layer can work independently, namely, the operation can be carried out and the result can be output as long as the input from the precursor node exists, and the input of all the precursor nodes is not required to be waited, the output result of the operation node is increased. Therefore, in the initial stage of the arrangement of the learning model, the target result can be quickly screened out due to the large data volume, and the weight of each operation node can be calculated more accurately. Over time, the learning model based on the invention will grow faster and more accurately than the conventional model, with the same amount of data and operation.
In addition, because the operation nodes of the invention have no strong dependence relationship, the application scene of the learning model of the invention is wider than that of the traditional model, the using environment can have various changing conditions, the operation nodes can be reused, special development and training for the specified scene are not needed, and the development cost is saved. In addition, when the user changes, the invention can also quickly adjust the weight of the operation node, thereby more quickly calculating the result meeting the user.
An embodiment of the search method applied to the project management software according to the present invention is described in detail below with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.
Fig. 1 is a flowchart of a search method applied to project management software according to an embodiment of the present invention. The present embodiment may be applied to a terminal operable by a user, as shown in fig. 1, the search method of the present embodiment includes:
step 101: the method comprises the step of accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently.
FIG. 2 is a diagram of a partial deployment of application project management software according to an embodiment of the present invention. As shown in fig. 2, in some embodiments, the learning model may be configured to a server in a cluster of log servers, and the server configured with the learning model is accessed to a terminal installed with project management software. When there are multiple servers, the learning model can be configured on a single server, and other servers can be configured in the server of the learning model, so that the learning model can access other servers. The user can send instruction data to the learning model through the terminal to start the search function of the project management software. The test machine cluster is connected with the terminal and the log server cluster, personnel can deploy specified software on a specified test machine in the test machine cluster through the terminal, and log data generated by the test machine can be sent to the log server cluster. It should be understood that the deployment diagram shown in fig. 2 is merely illustrative and is in no way intended to limit the invention, its application, or uses, as those skilled in the art will appreciate that it may be deployed reasonably according to actual needs.
Fig. 3 is a schematic diagram of a node architecture of a learning model according to an embodiment of the present invention. As shown in fig. 3, the neural structure search-based learning model includes an input layer, an output layer, and an operation layer. The input layer includes input nodes, the output layer includes output nodes, and the operational layer includes one or more operational nodes. The input nodes are collectively called input ends, and any input nodes capable of providing input data for the operation layer can be used as the input ends; the output nodes are collectively called output ends, and any terminal capable of receiving the result generated by the operation layer can be used as the output nodes; the input and output may be the same terminal. It should be understood that the node architecture shown in fig. 3 is illustrative only and is not intended to be in any way limiting of the invention and its applications or uses.
Although the node architecture of the learning model of the embodiment of the present invention is similar to the conventional model based on neural structure search, each operation node and its corresponding behavior are different. Specifically, in the conventional model, if one operation layer has a plurality of predecessor nodes, each operation node on the operation layer needs to wait for the outputs of all the predecessor nodes to reach the operation layer before starting to perform operation. However, in the searching method according to the embodiment of the present invention, the operation nodes on the operation layer are disassembled, so that each operation node of the operation layer can work in combination with other operation nodes or work alone, and thus the operation nodes can operate according to the outputs of some predecessor nodes without waiting until the outputs of all predecessor nodes reach the operation layer. For example, the operation node can separately execute the functions of analyzing the input condition and querying data.
Step 102: when the query condition is input into the project management software, the input node generates a search condition according to the query condition and sends the search condition to the operation node.
In some embodiments, when a user inputs a query condition to the project management software, the input node of the input layer may infer a usage scenario and generate a plurality of search conditions according to the query condition, and then generate a blank data set, store the search conditions in different data sets, and send the data sets to corresponding operation nodes, respectively. In some embodiments, the learning model may disassemble the query terms, for example, by type, into a range query term, a specified value query term, and a fuzzy query term. The input node presumes the usage scenario and generates a plurality of search conditions based on the query conditions.
Step 103: and the operation node performs operation according to the search condition, generates a data set and transmits the data set to the next operation node and the output node, wherein the data set comprises the weight of the operation node and the search result.
In the searching method of the embodiment of the invention, the operation node can directly operate the output of one or more precursor nodes including the input node, and the operation does not need to be started until the outputs of all the precursor nodes reach the operation layer. For example, the operation node can separately execute the functions of analyzing the input condition and querying data.
Moreover, in the conventional model based on NAS, an operation node can only output the optimized search condition to the next operation node, and the output of the operation node of the present invention is a set of data sets, which may include the search condition, the search result, the path and the weight of the operation node. In some embodiments, the data set of the compute node includes: (1) searching conditions used when the operation node is executed; (2) executing the search result obtained by the operation node; (3) the order in which the operation nodes receiving the output of the input node execute to the operation node (hereinafter referred to as the path of the operation node); (4) and executing the newly calculated weight value after the operation node is executed.
In some embodiments, when the operation node receives a data set of a previous node (the previous node may be an input node, and may also be an operation node, and this is referred to as a predecessor node in the following description), a search condition in the data set is decomposed and/or fused with a search condition that currently exists in the operation node, a new search condition is generated and stored in the data set of the operation node, and then the new search condition is used for performing an operation, and a search result is stored in the data set of the operation node. Optionally, the operation node may cover, by using a condition with a small query range, a condition with a wide query range, the search condition in the data set of the predecessor node and the existing search condition of the operation node according to the "keyword of query"; and fusing the principle of parallel storage if the inquired keywords are different to generate a new search condition.
In some embodiments, when the operation node receives the data set of the precursor node, the database of the server is searched by taking the search condition in the data set as a reference, and the searched data is saved as a search result in the data set.
In some embodiments, when an operation node receives a data set of a predecessor node, the operation node is attached to the back of the path of the predecessor node and is saved as the path of the operation node in the data set of the operation node.
In some embodiments, when the operation node receives the data set of the precursor node, the weight in the data set of the precursor node and the weight of the operation node may also be calculated to form a new weight and store the new weight in the data set of the operation node. Optionally, a calculation formula for calculating the weight in the data set of the predecessor node and the weight of the operation node is as follows:
the new weight = (the weight of the operation node ^ 2)/the weight in the data set of the precursor node + the weight in the data set of the precursor node.
After the search condition, the search result, the path and the weight of the operation node are obtained, the data set of the operation node is used as an output result to be transmitted to the next operation node and an output node in an output layer.
The method comprises the following steps: 104: and the next operation node repeatedly executes the same operation as the operation node until the learning model judges that the operation does not need to be continued.
In some embodiments, the next operation node performs the same operation as the operation node, and the description thereof is omitted here. And when the learning model judges that the data set of the operation node does not need to be transmitted to other operation nodes for continuous operation, only sending the data set of the operation node to the output layer.
In the operation layer of the embodiment of the invention, each operation node can directly perform operation according to the output of one or more precursor nodes and output a data set, and operation does not need to be started when the outputs of all the precursor nodes reach the operation layer, so that more output results than the number before modification can be obtained in the output layer.
Step 105: and the output node sorts the received search results in the data set according to the weight of the operation node, and sequentially outputs the search results to the terminal for display.
In some embodiments, the output nodes in the output layer sequence all the received data sets according to the ascending order of the weights of the operation nodes in the data sets, and output the search results in the data sets to the terminal in sequence so as to display the search results to users.
Based on the searching method provided by the embodiment of the invention, each operation section in the operation layer can directly perform operation according to the output of part of precursor nodes and output a data set, and operation is not required to be performed when the outputs of all the precursor nodes reach the operation layer, so that more output results than the number before modification can be obtained in the output layer. In the initial stage of the arrangement of the learning model in the embodiment of the invention, the target result can be quickly screened out due to the large data volume. Over time, the learning model of the embodiment of the invention will grow faster and more accurately than the conventional model with the same amount of data and operation.
Moreover, all the operation nodes in the learning model can work with other operation nodes in a combined mode and can work independently, and strong dependency relation does not exist among the operation nodes, so that the application scene of the learning model under the searching method is wider than that of the traditional model, the learning model is applicable to the condition that the environment has various changes, the operation nodes can be reused, special development and training for a specified scene are not needed, and the development cost is saved.
In practical application, the searching method can be used for maintaining software and searching infrequently, and under the use scene, the advantage of rapid growth of a learning model is far greater than the requirement of high accuracy at the initial operation stage.
Furthermore, the searching method can also modify the weight of the operation node in the learning model and/or modify the predecessor node and/or successor node of the operation node according to the selection of the user. For example, in the searching method of the present invention, the operation node may directly perform the operation according to some conditions without waiting for all conditions to be determined, and in this case, if the result of the operation performed by the missing condition for multiple times is the result intended by the user, it can be assumed that the missing condition is not important for the search, and thus the weight of the node on the path corresponding to the missing condition may be reduced.
Fig. 4 is a flowchart of a searching method applied to project management software according to another embodiment of the present invention. As shown in fig. 4, on the basis of the embodiment shown in fig. 1, the implementation of the present invention further includes step 106: when the user selects the output result with low weight value from the output results displayed by the terminal for multiple times, the weight value of each operation node in the learning model is recalculated, and/or the precursor node and/or the successor node of the operation node can be corrected.
In some optional examples, when the user selects the output result with a low weight value for multiple times, the steps of recalculating the weight value of the operation node and correcting the predecessor node and/or successor node of the operation node include:
step 1: and calculating the weights of all operation nodes on the path of the selected output result, and marking the operation nodes as the model soldier nodes. Optionally, for the model node, the weight calculation formula is:
and the operation node weight = the current weight X, which is an important value of the search condition.
Step 2: and calculating the weight values of the operation nodes contained in other output results of which the weight values are greater than the selected output result, and marking the operation nodes as non-standard soldier nodes. Optionally, for the non-standard soldier node, the weight calculation formula is:
operation node weight = current weight X the important value X12.73% of the current search condition X the number of times the operation node is used in the current output result.
And 3, step 3: and when the data set corresponding to the selected output result is not the first output result in the plurality of output results displayed by the terminal, calculating the weight of the model node again. Optionally, the formula for calculating the weight of the model node again is as follows:
operation node weight = current weight X ((order number X1.375% > +100% when data set of operation node is output)).
And 4, step 4: and taking the node of the operation node with the weight value larger than the minimum weight value in the model nodes in the non-model nodes as the successor node of the operation node with the minimum weight value. Optionally, when the number of the operation nodes of the whole path corresponding to the output result is greater than or equal to 8, the positions of the operation node with the minimum weight in the path are sequentially shifted out according to the sequence from the small value to the large value, that is, the predecessor node of the operation node with the minimum weight in the successor node in the path is changed into the predecessor node of the operation node with the minimum weight in the path, and the successor node of the operation node with the minimum weight in the predecessor node in the path is changed into the successor node of the operation node with the minimum weight in the path. In a specific embodiment, if the weights of the operation nodes on the path corresponding to the output result are 6, 7, 2, 3, 8, 5, and 8 in sequence, and the weight of the non-standard soldier node to be accessed to the path is 4, the path is changed to 6, 7, 2, 4, 3, 8, 5, and 8 after the access, at this time, the number of the operation nodes is 8, exceeds the maximum value, and the operation nodes are deleted in sequence from small to large according to the weight, and are changed to 6, 7, 4, 3, 8, 5, and 8.
In the searching method, the weight of the operation node and the predecessor node and successor node of the operation node can be continuously corrected along with the operation of searching, so that the weight of the operation node can be more accurately and intelligently calculated. Moreover, when the user changes, the invention can also quickly adjust the weight of the operation node, thereby calculating the result desired by the user more quickly.
In another embodiment of the present invention, the following steps may be further included: after the query is finished, the learning model is emptied. Optionally, after one search is finished, all data in the data set in each operation node in the learning model are deleted. Note that the weight data of the operation node is stored in the learning model as a node attribute.
In practical application, the searching method can be used for maintaining software and in the case of infrequent searching. Under the use scene, the advantage of the rapid growth of the learning model is far greater than the requirement of high accuracy in the initial operation stage.
The following further describes an application of the search method according to the embodiment of the present invention in a specific search scenario with reference to fig. 5. FIG. 5 is a schematic diagram of a workflow for searching log information according to an embodiment of the invention. Specifically, fig. 5 shows an embodiment of searching log data by a date condition, which shows a state based on that partial data already exists and a learning model has been trained for a while. In this embodiment, the learning model includes an input layer, an arithmetic layer, an output layer, and a model management module. As shown in fig. 5, the process of this search includes:
step A: the user enters the query conditions. Specifically, the query condition input by the user is "date =2022-6-6", and then the terminal transmits the query condition to the input node of the input layer in the learning model.
And B: the input node parses the query. Specifically, the input node disassembles the query conditions in step a into three search conditions of "date =2022-6-6", "information level = all", and "person = all". And then, storing the disassembled search conditions into a data set and sending the data set to an operation node which takes the date as the primary search condition in an operation layer. Step B corresponds to step 102 in fig. 1 and 2, and is not described herein again.
Step C: and the operation node analyzes the received data and performs subsequent processing. Specifically, as shown in fig. 4, the operation node receives the search conditions from the input node, queries the database on the server for data based on the search conditions, and saves the query result as a search result in the data set of the operation node. Meanwhile, the operation node calculates the weight of the results and stores the weight and the path calculated to the operation node into a data set. The process of generating the search data, the weight and the path corresponds to the process of generating the data set in step 103 in fig. 1 and 2, and is not described herein again.
The operation node divides the search condition into two groups of "date =2022-6-6", "information level = important", "person = whole person" and "date =2022-6-6", "information level = general", "person = whole person", and stores the search condition as a new search condition in each data set. And then, respectively sending the two groups of data sets to the subsequent nodes and the output layer.
And analyzing and inquiring data by the successive nodes until the search condition is not required to be disassembled continuously. Specifically, similar to step C, the subsequent node queries data in the database with the search condition in the received data set as a reference, and stores the query result as a search result in the data set of the operation node. Meanwhile, the successor node calculates the weight of the results, stores the weight and the path calculated to the successor node into a data set, and respectively sends the data set to the next node of the successor node and an output layer. And repeating the process until the next node analysis considers that the search condition is not required to be disassembled continuously, namely terminating the subsequent operation and query.
Step D: and the output nodes in the output layer sort the received data sets according to the weight value from large to small, and output the search results in the data sets as the presumed results to the terminal in sequence. Step D corresponds to step 105 in fig. 1 and 2, and is not described herein again.
Step E: the user selects the output result. Specifically, the user selects required data, and the terminal sends the data actually selected by the user to a model management module of the learning model, so that the learning model can recalculate the weight of each operation node, and/or correct predecessor and/or successor nodes of the operation node. Step E corresponds to step 106 in fig. 2, and is not described herein again.
FIG. 6 is a schematic structural diagram of a search apparatus applied to project management software according to an embodiment of the present invention. The searching apparatus of this embodiment can be used to implement the embodiments of the above-described searching methods of the present invention. As shown in fig. 6, the apparatus of this embodiment includes: an access module 601, a generation module 602, a first operation module 603, a second operation module 604 and an output module 605. Wherein:
the access module 601 is used for accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently;
a generating module 602, configured to, when a query condition is input to the project management software, enable the input node to generate a search condition according to the query condition, and send the search condition to the operation node;
a first operation module 603, configured to enable the operation node to perform operation according to the search condition, generate a data set, and transmit the data set to a next operation node and an output node, where the data set includes a weight of the operation node and a search result;
a second operation module 604, configured to enable the next operation node to repeatedly execute the same operation as the operation node until the learning model determines that no operation needs to be continued; and
and an output module 605, configured to enable the output node to sort the search results in the received data set according to the weight of the operation node, and output the search results to the terminal in sequence for display.
In addition to the above method and apparatus, an embodiment of the present invention further provides a computer device, where the computer device includes a processor and a memory, where the memory stores a computer program, and the computer program is loaded by the processor and executed to implement the search method according to any of the above embodiments of the present invention.
In addition, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored, and the computer program is loaded and executed by a processor to implement the search method according to any of the above-mentioned embodiments of the present invention.
In summary, in the searching method, apparatus, computing device and computer-readable storage medium of the present invention, since each operation node in the operation layer of the learning model can work independently, that is, as long as there is an input from a predecessor node, it can perform an operation and output a result without waiting for the inputs of all predecessor nodes, and thus the output results of the operation nodes are increased. Therefore, the target result can be quickly screened out. In addition, because the operation nodes do not have strong dependence relationship with each other, the learning model has wider application scene than the traditional model, the use environment can have various changing conditions, the operation nodes can be multiplexed, special development and training for the specified scene are not needed, and the development cost is saved.
In addition, the weight of the operation node and the predecessor node and successor node of the operation node can be continuously corrected along with the search operation, so that the weight of the operation node can be more accurately and intelligently calculated. When the user changes, the invention can also quickly adjust the weight of the operation node, thereby calculating the result required by the user more quickly.
Thus, the advantages of the present invention include at least:
1. the learning model of the invention has fast growth speed, can complete the training of the learning model by a small amount of data, and does not need to train the model by a large amount of data in advance, therefore, even if the times of using the search function is few, the user can be provided with the expected search result.
2. The invention can be suitable for the condition that the environment has various changes, and the study model does not need to be developed again aiming at different search scenes, thereby reducing the development cost and the maintenance cost of project management software.
3. Because the learning model of the invention can complete training by a small amount of data, when a user changes, the learning model can quickly calculate the search result according to the habit of the user.
Embodiments of the invention are operational with numerous other general purpose or special purpose computing system environments or configurations, and with numerous other electronic devices, such as terminal devices, computer systems, servers, etc. Examples of well known terminal devices, computing systems, environments, and/or configurations that may be suitable for use with the electronic devices described above include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above, and the like.
Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. The computer system/server may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.
While certain embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the invention, as the novel methods and systems described herein may be embodied in various other forms. Furthermore, various omissions, substitutions and changes in the form and method described herein may be made without departing from the spirit of the inventions. The accompanying claims and their equivalents are intended to cover such forms and modifications as would fall within the scope and spirit of the inventions.
Claims (10)
1. A search method applied to project management software is characterized by comprising the following steps:
accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently;
when a query condition is input into the project management software, the input node generates a search condition according to the query condition and sends the search condition to the operation node;
the operation node performs operation according to the search condition, generates a data set and transmits the data set to a next operation node and an output node, wherein the data set comprises the weight of the operation node and a search result;
the next operation node repeatedly executes the same operation as the operation node until the learning model judges that the operation does not need to be continued; and
and the output node sorts the received search results in the data set according to the weight of the operation node, and sequentially outputs the search results to the terminal for display.
2. The search method of claim 1, further comprising: and when the user selects the output result with low weight value from the output results displayed by the terminal for multiple times, modifying the weight value of the operation node in the learning model, and/or correcting the precursor node and/or the successor node of the operation node.
3. The search method according to claim 1 or 2, characterized in that the method further comprises: and after the query is finished, emptying temporary data in the learning model.
4. The method of claim 3, wherein the data set of the compute node comprises: the search condition of the operation node, the search result of the operation node, the path of the operation node and the weight of the operation node.
5. The searching method according to claim 4, wherein the weight values included in the data set of the operation nodes are the latest calculated weight values after the operation of the operation nodes is performed, and the calculation formula is as follows:
the new weight = (the weight of the operation node ^ 2)/the weight in the data set of the precursor node + the weight in the data set of the precursor node.
6. The searching method according to claim 4, wherein the steps of recalculating the weight of the operation node and correcting the predecessor node and/or successor node of the operation node when the user selects the output result having a low weight for a plurality of times comprise:
calculating the weights of all operation nodes on the path of the selected output result, marking the operation nodes as the model soldier nodes, and calculating the weights of the model soldier nodes by the following formula:
the operation node weight = the current weight X and the important value of the search condition;
calculating the weight values of the operation nodes contained in other output results of the selected output result, and marking the operation nodes as non-standard soldier nodes, wherein the weight value calculation formula of the non-standard soldier nodes is as follows:
operation node weight = current weight X the important value X12.73% of this search condition X the number of times this operation node uses in this output result;
when the data set corresponding to the selected output result is not the first output result in the plurality of output results displayed by the terminal, the weight of the model node is calculated again, and the weight calculation formula of the model node is calculated again as follows:
operation node weight = current weight X ((sequence number of output of data set of operation node X1.375%) + 100%)
And taking the node of the operation node with the weight value larger than the minimum weight value in the model nodes in the non-model nodes as the successor node of the operation node with the minimum weight value.
7. The search method of claim 3, wherein the search method is used in situations where software is maintained and searches are infrequent.
8. A search apparatus applied to project management software, comprising:
the access module is used for accessing a learning model based on neural structure search into project management software of a terminal, wherein the learning model comprises an input node, an output node and one or more operation nodes, and each operation node can work independently;
the first generation module is used for enabling the input node to generate a search condition according to the query condition when the query condition is input into the project management software and sending the search condition to the operation node;
the second generation module is used for enabling the operation node to operate according to the search condition, generating a data set and transmitting the data set to a next operation node and an output node, wherein the data set comprises a weight of the operation node and a search result;
the execution module is used for enabling the next operation node to repeatedly execute the same operation as the operation node until the learning model judges that the operation does not need to be continued; and
and the output module is used for enabling the output node to sort the search results in the received data set according to the weight of the operation node and sequentially output the search results to the terminal for displaying.
9. A computer device, characterized in that it comprises a processor and a memory, in which a computer program is stored, which computer program is loaded and executed by the processor to implement the search method according to any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored which is loaded and executed by a processor to implement the search method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310060044.2A CN115795125B (en) | 2023-01-18 | 2023-01-18 | Searching method, device, equipment and medium applied to project management software |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310060044.2A CN115795125B (en) | 2023-01-18 | 2023-01-18 | Searching method, device, equipment and medium applied to project management software |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115795125A true CN115795125A (en) | 2023-03-14 |
| CN115795125B CN115795125B (en) | 2023-04-18 |
Family
ID=85429760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310060044.2A Active CN115795125B (en) | 2023-01-18 | 2023-01-18 | Searching method, device, equipment and medium applied to project management software |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115795125B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111667057A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Method and apparatus for searching model structure |
| US20200302272A1 (en) * | 2019-03-19 | 2020-09-24 | Cisco Technology, Inc. | Systems and methods for auto machine learning and neural architecture search |
| CN112508178A (en) * | 2020-12-11 | 2021-03-16 | 鼎富智能科技有限公司 | Neural network structure searching method and device, electronic equipment and storage medium |
| CN113361578A (en) * | 2021-05-31 | 2021-09-07 | 北京百度网讯科技有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
| US20210312276A1 (en) * | 2020-04-07 | 2021-10-07 | International Business Machines Corporation | Automated decision making for neural architecture search |
-
2023
- 2023-01-18 CN CN202310060044.2A patent/CN115795125B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200302272A1 (en) * | 2019-03-19 | 2020-09-24 | Cisco Technology, Inc. | Systems and methods for auto machine learning and neural architecture search |
| US20210312276A1 (en) * | 2020-04-07 | 2021-10-07 | International Business Machines Corporation | Automated decision making for neural architecture search |
| CN111667057A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Method and apparatus for searching model structure |
| CN112508178A (en) * | 2020-12-11 | 2021-03-16 | 鼎富智能科技有限公司 | Neural network structure searching method and device, electronic equipment and storage medium |
| CN113361578A (en) * | 2021-05-31 | 2021-09-07 | 北京百度网讯科技有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115795125B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10380498B1 (en) | Platform services to enable one-click execution of the end-to-end sequence of modeling steps | |
| JP5147162B2 (en) | Method and system for determining object similarity based on heterogeneous relationships | |
| CN102193929B (en) | Method and equipment for searching by using word information entropy | |
| Hatzi et al. | The PORSCE II framework: Using AI planning for automated semantic web service composition | |
| US9418107B2 (en) | Method and apparatus for performing query aware partitioning | |
| US10963301B2 (en) | Scheduling operations on a computation graph | |
| CN101719932B (en) | System and method for automatic service combination | |
| CN112070416B (en) | AI-based RPA flow generation method, apparatus, device and medium | |
| CN111881345B (en) | Neural collaborative filtering service quality prediction method based on position context awareness | |
| CN110119399B (en) | Business process optimization method based on machine learning | |
| CN115017159A (en) | Data processing method and device, storage medium and electronic equipment | |
| CN114817389A (en) | Data processing method, data processing device, storage medium and electronic equipment | |
| CN115795125B (en) | Searching method, device, equipment and medium applied to project management software | |
| Chen et al. | Efficient distributed clustering algorithms on star-schema heterogeneous graphs | |
| US20150081672A1 (en) | Automatic conversion of units of measure during data stream processing | |
| CN111464451A (en) | Data stream equivalent connection optimization method and system and electronic equipment | |
| US20240005170A1 (en) | Recommendation method, apparatus, electronic device, and storage medium | |
| WO2022252694A1 (en) | Neural network optimization method and apparatus | |
| US20230195742A1 (en) | Time series prediction method for graph structure data | |
| CN112070487B (en) | AI-based RPA flow generation method, apparatus, device and medium | |
| JP2023002103A (en) | Model inference device, method, and program | |
| CN110210691B (en) | Resource recommendation method, device, storage medium and equipment | |
| CN111400399A (en) | Account book synchronization method and device of block chain system and hardware equipment | |
| Paulus et al. | Filtering Recommender System for Semantic Model Refinement | |
| CN114861057B (en) | Resource sending method, training of recommendation model and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |