Multi-source heterogeneous college database-oriented information indexing method and electronic equipment

Abstract

The invention discloses an information indexing method for a multi-source heterogeneous college database, electronic equipment and a storage medium, and the information indexing method for the multi-source heterogeneous college database comprises the following steps: constructing a metadata vector set based on the database and a word vector frequent set of multiple pieces of historical retrieval information; and determining a target candidate item in the metadata vector set, then analyzing the target candidate item based on a knowledge graph triple relationship and a TransE model to obtain a preference feature vector, and then analyzing a preliminary retrieval result based on a bidirectional recurrent neural network model and the preference feature vector to obtain a to-be-retrieved resource. According to the information indexing method for the multi-source heterogeneous college database, the retrieved information can be obtained more accurately and efficiently.

Description

Technical Field

[0001] The present application relates to the field of information retrieval technology, and in particular to an information indexing method and electronic equipment for multi-source heterogeneous university databases. Background Art

[0002] Driven by the rapid development of innovation, a large number of intellectual property information resources have been collected at high speed. Resource users can obtain the content of intellectual property information resources through the Internet, which facilitates the full use of resources by demanders and protects the interests of intellectual property owners of resources. However, with the increase in the number of optional intellectual property information resources, it has also brought great difficulties for users to use resources accurately to avoid infringement.

[0003] A paper titled "Application of Natural Language Processing Technology in Educational Resources Retrieval" published in the Journal of Physics proposed an information retrieval method that uses natural language processing technology to achieve good processing effects on text information resources, and improves retrieval performance by simplifying the matching relationship between retrieval keywords and resources. However, this retrieval method can only achieve better retrieval results when applied to small databases, and the effect is poor when applied to heterogeneous and large-volume resource retrieval. Summary of the invention

[0004] The embodiments of the present application provide an information indexing method, electronic device and storage medium for multi-source heterogeneous university databases to solve the problem of poor retrieval effect of heterogeneous and large-scale resources in the prior art.

[0005] In a first aspect, an embodiment of the present application provides an information indexing method for a multi-source heterogeneous university database, including:

[0006] Reconstruct the metadata vector set based on various heterogeneous databases, and each element in the metadata vector set is represented by a word vector;

[0007] Obtain multiple historical search information and construct frequent word vector sets for multiple historical search information;

[0008] Based on the frequent word vector set, determine the target candidate items in the metadata vector set;

[0009] Analyze the target candidate items based on the knowledge graph triple relationship and TransE model to obtain the preference feature vector;

[0010] Obtaining target search information and its corresponding preliminary search results in the metadata vector set;

[0011] The preliminary search results are analyzed based on the bidirectional recurrent neural network model and preference feature vector to obtain the resources to be searched.

[0012] In a possible implementation of the first aspect, analyzing the preliminary search results based on the bidirectional recurrent neural network model and the preference feature vector to obtain the resources to be searched includes:

[0013] Determine the resources to be searched based on the preliminary search results;

[0014] Based on the bidirectional recurrent neural network model and preference feature vector, each resource to be searched is sorted and output.

[0015] In a possible implementation of the first aspect, the resources to be searched are sorted and output based on the bidirectional recurrent neural network model and the preference feature vector, including:

[0016] For each retrieval resource in the preliminary retrieval results, the similarity weight between the retrieval resource and the target retrieval information is obtained based on the bidirectional recurrent neural network model, and the attention value of the retrieval resource is determined based on the similarity weight;

[0017] Obtain all search resources whose attention values ​​are higher than the set value as resources to be searched;

[0018] Obtaining the similarity between each resource to be retrieved and the preference feature vector;

[0019] The resources to be detected are sorted in order from high to low according to their similarity and then output.

[0020] In a possible implementation of the first aspect, obtaining the similarity between each to-be-searched resource and the preference feature vector includes:

[0021] Obtaining the matching degree between each resource to be searched and the target search information;

[0022] Arrange the resources to be searched in descending order according to the matching degree;

[0023] Select some of the top-ranked resources to be searched and obtain the similarity between them and the preferred feature vector;

[0024] According to the similarity from high to low, the top ranked resources to be searched are sorted in sequence and then output.

[0025] In a possible implementation of the first aspect above, the matching degree is obtained based on a TF-IDF algorithm.

[0026] In a possible implementation of the first aspect above, the attention value of the retrieved resource is obtained based on a bidirectional recurrent neural network model including a gated recurrent unit.

[0027] In a possible implementation of the first aspect, determining a target candidate item in a metadata vector set based on a frequent word vector set includes:

[0028] Based on the frequent word vector set, all resources associated with multiple historical search information in the metadata vector set are obtained as candidate item sets;

[0029] Get the distance between each element in the candidate item set and the frequent word vector set as the support of the element;

[0030] The target candidate item is determined based on the element with the smallest support.

[0031] In a possible implementation of the first aspect above, the target candidate item satisfies a set retrieval association rule.

[0032] In a possible implementation of the first aspect, reconstructing the metadata vector set based on the heterogeneous databases includes:

[0033] Obtain metadata from various heterogeneous databases;

[0034] The metadata is standardized, deduplicated and integrated according to the RDF format standard using conversion technology;

[0035] Based on the Skip-gram word vector calculation method, the processed metadata are converted into word vector form.

[0036] In a second aspect, an electronic device is provided in an implementation of the present application, comprising: a memory for storing instructions executed by one or more processors of the electronic device; and a processor, which, when the processor executes the instructions in the memory, enables the electronic device to execute a method in any possible implementation of the above-mentioned first aspect.

[0037] In a third aspect, the present application provides a computer-readable storage medium on which instructions are stored, and when the instructions are executed on a computer, the computer executes a method in any possible implementation of the first aspect.

[0038] Compared with the prior art, this application has at least the following technical effects:

[0039] After preliminary processing of each resource data in each heterogeneous database, the knowledge graph triple relationship and TransE model are further used to mine the user's deep consciousness when searching for resources, and the user's preference feature vector for the searched resources is obtained. Then, the user's preference feature vector for the searched resources is used as the last link of resource matching to complete the search of information resources, and the search results are fed back to the user in order. This application uses the data mining-based method proposed in this application to obtain the retrieved information more accurately and efficiently, and when this method is applied to fuzzy resource retrieval, the retrieved resources are more in line with the searcher's demand standards, and its retrieval effect is greatly improved compared to the current resource retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 According to some embodiments of the present application, a schematic diagram of the process of an information indexing method for a multi-source heterogeneous university database is shown. Figure 1 ;

[0041] Figure 2 According to some embodiments of the present application, a schematic flow chart of a method for constructing a metadata vector set is shown;

[0042] Figure 3 According to some embodiments of the present application, a schematic diagram of the process of an information indexing method for a multi-source heterogeneous university database is shown. Figure 2 ;

[0043] Figure 4 According to some embodiments of the present application, a distribution diagram of matching values ​​between the retrieval output of each retrieval method and the predicted retrieval results is shown;

[0044] Figure 5 According to some embodiments of the present application, a schematic diagram of the structure of an electronic device is shown;

[0045] Figure 6 According to some embodiments of the present application, a schematic structural diagram of a system on chip (SOC) is shown. DETAILED DESCRIPTION

[0046] The present application is further described below in conjunction with specific embodiments and drawings. It is understood that the illustrative embodiments of the present disclosure include, but are not limited to, information indexing methods, electronic devices, and storage media for multi-source heterogeneous university databases. The specific embodiments described herein are merely for the purpose of explaining the present application, not for limiting the present application. In addition, for ease of description, only some, but not all, structures or processes related to the present application are shown in the drawings.

[0047] The following specific embodiments illustrate the implementation of the present application, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Although the description of the present application will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this application are limited to the implementation. On the contrary, the purpose of introducing the invention in conjunction with the implementation is to cover other options or modifications that may be extended based on the claims of the present application. In order to provide a deep understanding of the present application, the following description will include many specific details. The present application can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present application, some specific details will be omitted in the description. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0048] With the rapid promotion of innovation and development, people's awareness of intellectual property protection is constantly increasing. As one of the important scientific research centers, how to effectively utilize intellectual property information resources and improve the efficiency of using intellectual property information resources is one of the main issues that universities cannot avoid in conducting teaching and scientific research.

[0049] The types of information resources stored in the university intellectual property information resource library are relatively rich, and it is impossible to use a metadata standard to uniformly describe all types of intellectual property information resources. When managing intellectual property information resource data, different resources are usually stored in different and independent heterogeneous databases for the convenience of management, which also brings certain difficulties to university information retrieval.

[0050] Since data mining technology can deeply explore the deep associations in intellectual property information resource data, it increases the matching success rate between the search text and the correct university intellectual property information resources, thereby improving resource retrieval efficiency. Based on this, an embodiment of the present application provides an information indexing method for multi-source heterogeneous university databases, including: constructing a metadata vector set based on each heterogeneous database and a frequent word vector set of multiple historical search information, based on the frequent word vector set, determining the target candidate item in the metadata vector set, and then analyzing the target candidate item based on the knowledge graph triple relationship and the TransE model to obtain the preference feature vector, and then further obtaining the target search information and its corresponding preliminary search results in the metadata vector set, analyzing the preliminary search results based on the bidirectional recurrent neural network model and the preference feature vector to obtain the resources to be searched.

[0051] After preliminary processing of each resource data in each heterogeneous database, this application further uses the knowledge graph triple relationship and TransE model to mine the user's deep consciousness when searching for resources, obtains the user's preference feature vector for the searched resources, and then uses the user's preference feature vector for the searched resources as the last link in resource matching to complete the search for information resources and feed back the search results to the user in order. The data mining-based method proposed in this application can more accurately and efficiently obtain the retrieved information, and when this method is applied to fuzzy resource retrieval, the retrieved resources are more in line with the searcher's demand standards, and the retrieval effect is greatly improved compared to the current resource retrieval.

[0052] The embodiments of the present application will be described in further detail below in conjunction with the accompanying drawings.

[0053] Figure 1 FIG. 1 is a flowchart of an information indexing method for a multi-source heterogeneous university database according to an embodiment of the present application. Figure 1 As shown, the information indexing method for multi-source heterogeneous university databases includes the following steps:

[0054] Step S1: Reconstruct the metadata vector set based on each heterogeneous database, and each element in the metadata vector set is represented by a word vector.

[0055] Specifically, Figure 2 As shown, the step includes:

[0056] S101, obtaining metadata from various heterogeneous databases.

[0057] S102, the metadata is standardized, deduplicated and integrated according to the RDF format standard using conversion technology.

[0058] S103, converting the processed metadata into word vector form based on the Skip-gram word vector calculation method.

[0059] Furthermore, since the types of information resources stored in the intellectual property information resource library of universities are relatively rich and of different types, different resources are generally stored in different and independent databases for the convenience of management. Therefore, it is necessary to integrate these heterogeneous intellectual property information resources of universities and establish a mapping relationship between resource metadata and search text. In order to realize the conversion of the input search request into the resource library search instruction corresponding to each heterogeneous university intellectual property information resource library during the search, and finally return the search results according to the resource mapping relationship of the metadata. That is, it is necessary to reconstruct and integrate the metadata of university intellectual property information resources.

[0060] When reconstructing and integrating metadata of university intellectual property information resources, the resources input by the management side are generally described according to the original data specifications of the input database; for external application resources, the metadata of new resources is mainly obtained by harvesting; for university intellectual property information resources publicly released on the Internet, the WEB crawler method is used to obtain open metadata by analyzing the URL.

[0061] After obtaining the metadata of all university intellectual property information resources, it is also necessary to standardize, deduplicate and integrate the metadata in a unified manner according to the RDF format standards using XML and other conversion technologies.

[0062] Considering that the search vectors often used in searching for intellectual property information resources are mainly in text form, this application performs text vectorization representation processing on the intellectual property information resource data. After conversion to metadata, in order to improve the matching success rate between the intellectual property information resources of universities and the corresponding search vectors during retrieval and reduce the influence of other texts in the information resources, a denoising model is used to remove noise from the resource information. This application filters the information with low relevance to the subject in the intellectual property information resources of universities, and uses the regularization principle to extract stems from text words and sentences to reduce the interference of irrelevant information. At the same time, according to the original language grammatical relationship in the information resource text, the text spaces are segmented, the words are segmented, and the stop words in the text are removed.

[0063] This application uses the Skip-gram word vector model to vectorize the resource metadata text in the university intellectual property information resource library. The Skip-gram word vector calculation model predicts all possible output word vectors based on the input information resource text, and selects the word vector with the maximum probability as the conversion result. The calculation formula of the Skip-gram word vector calculation model is as follows:

[0064]

[0065] Among them, W(r) is the input university intellectual property information resource text; P is the corresponding word vector e of the predicted output k The probability of ; E is the word vector matrix of university intellectual property information resources text; W is the weight matrix of word vectorization processing; S is the output of the Skip-gram word vector calculation model; N is the number of resource text words to be processed.

[0066] Step S2: Obtain multiple historical search information and construct frequent word vector sets of the multiple historical search information.

[0067] When users search for university intellectual property information resources, all the search term vectors they input contain information such as the user's search resource awareness and the user's preference for resource types. Therefore, we can combine the correlation between the search term vectors entered by users when searching for resources to deeply explore the user's search awareness and improve the accuracy of the search.

[0068] Specifically, according to the processing process of information resource data in each heterogeneous database in step S1, each historical search information input by the user during search can be segmented and knowledge point mapped, and all structural knowledge points associated with the user's search information knowledge points in the resource database can be traversed to construct a vector frequent set of historical search information.

[0069] Step S3: Based on the frequent word vector set, determine the target candidate item in the metadata vector set.

[0070] Specifically, step S3 may specifically include: based on the word vector frequent set, obtaining all resources associated with multiple historical retrieval information in the metadata vector set as a candidate item set; obtaining the distance between each element in the candidate item set and the word vector frequent set as the support of the element; and determining the target candidate item based on the element with the smallest support.

[0071] Furthermore, the target candidate items need to satisfy the set retrieval association rules.

[0072] Step S4: Analyze the target candidate items based on the knowledge graph triple relationship and TransE model to obtain the preference feature vector.

[0073] Specifically, according to the triple relationship of the knowledge graph, the TransE model is used to analyze the association between knowledge points and user search awareness. For the knowledge graph triple in the user search information, the objective function of the TransE model that represents the relationship between two knowledge point entities is as follows:

[0074]

[0075] Among them, χ is the partition coefficient of the association relationship between two knowledge point entities; f(c, r, g) is the scoring function of the TransE model. The association rules between the user's search awareness and the intellectual property information resources of the universities to be searched are obtained from the knowledge graph triples in the search information, and the user's resource search preference, i.e., search awareness, is mined using the data mining algorithm based on association rules.

[0076] Step S5: Obtain target search information and its corresponding preliminary search results in the metadata vector set.

[0077] Step S6: Analyze the preliminary search results based on the bidirectional recurrent neural network model and the preference feature vector to obtain the resources to be searched.

[0078] Specifically, Figure 3 As shown, step S6 specifically includes:

[0079] Step S601: For each search resource in the preliminary search results, obtain the similarity weight between the search resource and the target search information, and determine the attention value of the search resource based on the similarity weight.

[0080] Furthermore, in the bidirectional recurrent neural network, a gated recurrent unit is added to improve the training efficiency of the network. The output of the gated recurrent unit is t as follows:

[0081]

[0082]

[0083] Among them, x t is the sample size of the input bidirectional recurrent neural network; b t represents the update gate of the gated recurrent unit; v t is the reset gate of the gated recurrent unit; ω is the weight coefficient of the gated recurrent unit.

[0084] In order to avoid large retrieval bias due to small number of matches when searching for university intellectual property information resources, adding an attention mechanism to the retrieval structure of the bidirectional recurrent neural network can facilitate in-depth understanding of the full text of the initially determined retrieval resources.

[0085] In the self-attention mechanism model, the similarity between the user's search information and the preliminary resource search results is first used to obtain the similarity weight. After normalizing the similarity weight, the search information and resource similarity weights are weighted and summed to obtain the resource's attention value:

[0086]

[0087] Among them, α j is the similarity weight after normalization of the softmax function; K j is the value entered by the user for the search vector; M is the number of resources to be processed.

[0088] Step S602: Acquire all search resources whose attention values ​​are higher than a set value as resources to be searched.

[0089] Step S603: Obtain the similarity between each resource to be searched and the preference feature vector.

[0090] Step S604: sort the resources to be detected in descending order according to their similarity and then output them.

[0091] Furthermore, the similarity between each resource to be searched and the preference feature vector is obtained, and each resource to be detected is sorted in order from high to low according to the similarity and then output, including the following steps: based on the TF-IDF algorithm, the matching degree between each resource to be searched and the target search information is obtained; each resource to be searched is sorted in descending order according to the matching degree; some of the resources to be searched with the highest ranking are selected to obtain the similarity between them and the preference feature vector; some of the resources to be searched with the highest ranking are sorted in order from high to low according to the similarity and then output.

[0092] Specifically, the TF-IDF algorithm model is:

[0093]

[0094] Among them, U k is the number of times the search term appears in the resources to be searched; maxU is the number of times the most frequently repeated term appears in the university intellectual property information resources; n is the number of all university intellectual property information resources in the search target library; n k The number of resources that contain the search term.

[0095] The following is a test to evaluate the performance of the information indexing method for multi-source heterogeneous university databases proposed in the embodiment of the present application, and to compare it with other existing retrieval methods. The experiment is completed in the university intellectual property information resource library and the retrieval effect is analyzed.

[0096]

Experimental preparation

[0097] This experiment uses a university's intellectual property information resource database as the experimental data source. The total resource type, resource content, resource classification label and other detailed information of the database are known. A certain amount of intellectual property information resource data is extracted from the database to form a training set, which is used to assist the retrieval method in determining various basic retrieval parameters.

[0098] After extracting the resource data constituting the training set from the university intellectual property information resource database, different amounts of resource data were randomly extracted from the intellectual property information resource database, and after manual processing, 10 groups of retrieval experimental subsets were formed to test the performance of the resource deceleration method. The manual processing of the experimental resource data set mainly includes adding interference resource data with high similarity and changing the number of interference resources in the experimental data set. The relevant parameters of the intellectual property information resource experimental set actually used in this experimental study are shown in Table 1 below.

[0099] Table 1 Parameters of the experimental dataset of intellectual property information resources

[0100]

[0101] The original data of intellectual property information resources used in this experiment do not overlap with each other in each data subset to avoid interference with the retrieval method experiment. When adding interference resources with different proportions and high similarity to the original resource data, manual annotation is performed to facilitate the subsequent calculation of the retrieval accuracy of the resource retrieval method.

[0102]

Experimental process

[0103] The purpose of the experiment is to verify the applicability of the information indexing method for multi-source heterogeneous university databases of the embodiment of the present application by intuitively and scientifically acquiring and analyzing the experimental indicator data. Therefore, the specific completion form of the experiment is to compare the method of the present application with the resource retrieval method based on natural language processing technology and the resource retrieval method based on deep learning. Under the comparison of the three resource retrieval methods, by analyzing the data values ​​of each retrieval method under the experimental verification indicators, the effect of the retrieval method in actual application to the retrieval of intellectual property information resources is analyzed.

[0104] The experiment selected the F1 value and the complete match value as the test indicators of the retrieval effect of the retrieval method, and selected the retrieval time consumption indicator of the method when retrieving the target resource as a supplement to achieve a comprehensive performance evaluation of the resource retrieval method. Among them, the complete match value is the proportion of the number of times the retrieval results of the retrieval method are completely consistent with the expected retrieval results to the total number of retrievals. The F1 value can represent the average overlap between the predicted resource retrieval results and the actual retrieval results, that is, it is calculated using the retrieval accuracy and recall rate of the retrieval method.

[0105] According to the set search labels, the selected search methods are used on the prepared experimental data subsets to complete the corresponding search work. The same search label is searched 10 times, and the average value is taken as the search result of the current label. A total of 50 intellectual property information resource searches under different search requirements are carried out on each experimental subset, and the correct search times, incorrect search times and other data of each method on the current experimental subset are recorded. After the statistical data is calculated and processed using MAATLAB 2012a software, the specific values ​​of the corresponding experimental indicators are obtained. The experimental indicator data obtained after processing are analyzed horizontally and vertically to evaluate the performance of each selected search method.

[0106]

Experimental results

[0107] The following three aspects are used to verify the search results of each search method.

[0108] (1) When using different search methods to search for intellectual property information resources, the number of correct searches of the search methods was counted, and the search accuracy and recall rate corresponding to each method were calculated. The F1 value shown in Table 2 was obtained after calculation.

[0109] Table 2 Comparison of F1 values ​​for retrieval of intellectual property information resources

[0110]

[0111]

[0112] From the analysis of the F1 values ​​of each resource retrieval method in Table 2 above, it can be seen that when searching for experimental data subsets with different parameters according to the retrieval requirements, the information indexing method for multi-source heterogeneous university databases in this application has a higher F1 value than the other two retrieval methods in each experimental group. From the perspective of the difference between the maximum and minimum F1 values ​​of the retrieval methods, the maximum and minimum F1 values ​​of the information indexing method for multi-source heterogeneous university databases in this application differ by about 4.6; the maximum and minimum F1 values ​​of the method based on natural language processing technology differ by about 11.8; the maximum and minimum F1 values ​​of the resource retrieval method based on deep learning differ by about 18.6.

[0113] The above data show that the information indexing method for multi-source heterogeneous university databases in this application has a more stable retrieval effect and higher reliability.

[0114] (2) When searching for the corresponding university intellectual property information resources, the best search results are pre-set, and the search outputs of each search method are matched with the predicted search results. According to the definition of the complete match value, the complete match value of the search method used in the experiment is calculated and obtained. Figure 4 The comparison results.

[0115] analyze Figure 4 It can be seen from the data in that when searching for intellectual property information resources, the information indexing method for multi-source heterogeneous university databases of this application has a complete match value curve that is always in the range of 90-95%. And when processing experimental resource data sets with different parameters, the complete match value of the retrieval method fluctuates less. The complete match values ​​of the resource retrieval method based on natural language processing technology vary greatly depending on different experimental data subsets, resulting in severe jitter in the complete match value curve of the method. Overall, the complete match value curve of the resource retrieval method based on deep learning is located in the middle part of the above two retrieval methods. From the perspective of the complete match value, the information indexing method for multi-source heterogeneous university databases of this application has a retrieval accuracy rate of more than 90%, and the retrieval effect is better than the other two retrieval comparison methods.

[0116] (3) According to the set retrieval input vector, when using different retrieval methods to retrieve resources on the experimental data set, the retrieval time statistics of each retrieval method on the experimental data subset are shown in Table 3 below.

[0117] Table 3 Average retrieval time of retrieval methods on different data sets (ms)

[0118]

[0119] From the analysis of the retrieval time data of each retrieval method in different experimental data subsets, it can be seen that the retrieval time of the information indexing method for multi-source heterogeneous university databases of this application is much less than that of the other two resource retrieval methods. The data in Table 3 are further averaged. In this experiment, the average retrieval time of the information indexing method for multi-source heterogeneous university databases of this application is 150.19ms, the average retrieval time of the resource retrieval method based on natural language processing technology is 455.02ms, and the average retrieval time of the resource retrieval method based on deep learning is 269.67ms. When using the information indexing method for multi-source heterogeneous university databases of this application to search for intellectual property information resources, the retrieval time is shortened by at least about 44.31%, and the retrieval efficiency is greatly improved.

[0120] Summarizing the above experimental indicator analysis, it can be seen that when providing intellectual property information resource retrieval services for universities, the information indexing method proposed in this application for multi-source heterogeneous university databases can obtain the retrieved information more accurately and efficiently, and when this method is applied to fuzzy resource retrieval, the retrieved resources are more in line with the searcher's demand standards, and the retrieval effect is greatly improved compared to the current resource retrieval.

[0121] Reference now Figure 5, shown is a block diagram of an electronic device 400 according to one embodiment of the present application. The electronic device 400 may include one or more processors 401 coupled to a controller hub 403. For at least one embodiment, the controller hub 403 communicates with the processor 401 via a multi-drop bus such as a front side bus (FSB), a point-to-point interface such as a QuickPath Interconnect (QPI), or a similar connection 406. The processor 401 executes instructions that control general types of data processing operations. In one embodiment, the controller hub 403 includes, but is not limited to, a graphics & memory controller hub (GMCH) (not shown) and an input / output hub (IOH) (which can be on a separate chip) (not shown), wherein the GMCH includes a memory and a graphics controller and is coupled to the IOH.

[0122] The electronic device 400 may also include a coprocessor 402 and a memory 404 coupled to a controller hub 403. Alternatively, one or both of the memory and the GMCH may be integrated within the processor (as described in this application), with the memory 404 and the coprocessor 402 being directly coupled to the processor 401 and the controller hub 403, with the controller hub 403 being in a single chip with the IOH.

[0123] The memory 404 may be, for example, a dynamic random access memory (DRAM), a phase change memory (PCM), or a combination of the two. The memory 404 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. The computer-readable storage medium stores instructions, specifically, temporary and permanent copies of the instructions. The instructions may include: when executed by at least one of the processors, causing the electronic device 400 to implement the following: Figures 1 to 3 When the instructions are executed on a computer, the computer is caused to execute the method disclosed in the above embodiment.

[0124] In one embodiment, coprocessor 402 is a special purpose processor, such as, for example, a high throughput Many Integrated Core (MIC) processor, a network or communication processor, a compression engine, a graphics processor, a general-purpose computing on graphics processing units (GPGPU), or an embedded processor. The optional nature of coprocessor 402 is indicated by a dashed line in FIG. Figure 5 middle.

[0125] In one embodiment, the electronic device 400 may further include a network interface (Network Interface Controller, NIC) 406. The network interface 406 may include a transceiver for providing a radio interface for the electronic device 400, thereby communicating with any other suitable device (such as a front-end module, an antenna, etc.). In various embodiments, the network interface 406 may be integrated with other components of the electronic device 400. The network interface 406 may implement the functions of the communication unit in the above-mentioned embodiments.

[0126] The electronic device 400 may further include an input / output (I / O) device 405. The I / O 405 may include: a user interface designed to enable a user to interact with the electronic device 400; a peripheral component interface designed to enable peripheral components to interact with the electronic device 400; and / or a sensor designed to determine environmental conditions and / or location information related to the electronic device 400.

[0127] It is worth noting that Figure 5 This is for illustrative purposes only. Figure 5 It is shown that the electronic device 400 includes multiple devices such as a processor 401, a controller hub 403, a memory 404, etc.; however, in actual applications, the device using the methods of the present application may include only a portion of the devices in the electronic device 400, for example, it may include only the processor 401 and the network interface 406. Figure 5 The properties of optional devices are shown with dashed lines.

[0128] Reference now Figure 6 , which is a block diagram of a system on chip (SoC) 500 according to an embodiment of the present application. Figure 6 In the FIG, similar components have the same reference numerals. In addition, the dashed boxes are optional features of more advanced SoCs. Figure 6In the embodiment, SoC 500 includes: an interconnect unit 550, which is coupled to a processor 510; a system agent unit 580; a bus controller unit 590; an integrated memory controller unit 540; a group or one or more coprocessors 520, which may include integrated graphics logic, an image processor, an audio processor, and a video processor; a static random access memory (SRAM) unit 530; and a direct memory access (DMA) unit 560. In one embodiment, the coprocessor 520 includes a special-purpose processor, such as, for example, a network or communication processor, a compression engine, a general-purpose computing on graphics processing units (GPGPU), a high-throughput MIC processor, or an embedded processor.

[0129] Static random access memory (SRAM) unit 530 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. The computer-readable storage medium stores instructions, and more specifically, stores temporary and permanent copies of the instructions. The instructions may include: when executed by at least one of the processors, causing the SoC to implement the following: Figures 1 to 3 When the instructions are executed on a computer, the computer is caused to execute the method disclosed in the above embodiments.

[0130] Each method implementation mode of the present application can be implemented in the form of software, magnetic components, firmware, etc.

[0131] Program code can be applied to input instructions to perform the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, a processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0132] Program code can be implemented with high-level programming language or object-oriented programming language to communicate with the processing system. When necessary, program code can also be implemented with assembly language or machine language. In fact, the mechanism described in this application is not limited to the scope of any specific programming language. In either case, the language can be a compiled language or an interpreted language.

[0133] One or more aspects of at least one embodiment may be implemented by representative instructions stored on a computer-readable storage medium, which represent various logics in a processor, and when the instructions are read by a machine, the machine makes logic for executing the technology of the present application. These representations, known as "Intellectual Property (IP) cores," may be stored on tangible computer-readable storage media and provided to multiple customers or production facilities to be loaded into a manufacturing machine that actually manufactures the logic or processor.

[0134] In some cases, the instruction converter can be used to convert instructions from a source instruction set to a target instruction set. For example, the instruction converter can transform (e.g., using static binary transformation, dynamic binary transformation including dynamic compilation), deform, simulate, or otherwise convert instructions into one or more other instructions to be processed by the core. The instruction converter can be implemented in software, hardware, firmware, or a combination thereof. The instruction converter can be on the processor, outside the processor, or partially on the processor and partially outside the processor.

[0135] In the accompanying drawings, some structural or method features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be required. Instead, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. In addition, the inclusion of structural or method features in a particular figure does not mean that such features are required in all embodiments, and in some embodiments, these features may not be included or may be combined with other features.

[0136] It should be noted that the units / modules mentioned in the various device embodiments of the present application are all logical units / modules. Physically, a logical unit / module can be a physical unit / module, or a part of a physical unit / module, or can be implemented as a combination of multiple physical units / modules. The physical implementation method of these logical units / modules themselves is not the most important. The combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed by the present application. In addition, in order to highlight the innovative part of the present application, the above-mentioned device embodiments of the present application do not introduce units / modules that are not closely related to solving the technical problems proposed by the present application, which does not mean that there are no other units / modules in the above-mentioned device embodiments.

[0137] It should be noted that in the examples and description of this patent, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0138] Although the present application has been illustrated and described with reference to certain preferred embodiments thereof, it will be apparent to those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present application.

Claims

1. An information indexing method for multi-source heterogeneous university databases, used to retrieve information in heterogeneous databases, characterized in that: include: Reconstructing a metadata vector set based on the database, wherein each element in the metadata vector set is represented by a word vector; Acquire multiple historical search information and construct frequent word vector sets of the multiple historical search information; Determining a target candidate item in the metadata vector set based on the frequent word vector set; Analyze the target candidate items based on the knowledge graph triple relationship and the TransE model to obtain a preference feature vector; Obtaining target search information and its corresponding preliminary search results in the metadata vector set; The preliminary search results are analyzed based on a bidirectional recurrent neural network model and the preference feature vector to obtain resources to be searched.

2. The information indexing method for multi-source heterogeneous university databases as claimed in claim 1 is characterized in that: The preliminary search results are analyzed based on the bidirectional recurrent neural network model and the preference feature vector to obtain resources to be searched, including: Determine each resource to be searched based on the preliminary search results; Based on the bidirectional recurrent neural network model and the preference feature vector, each of the resources to be searched is sorted and output.

3. The information indexing method for multi-source heterogeneous university databases as claimed in claim 2 is characterized in that: The method outputs the sorted resources to be searched based on the bidirectional recurrent neural network model and the preference feature vector, including: For each search resource in the preliminary search results, obtaining a similarity weight between the search resource and the target search information based on the bidirectional recurrent neural network model, and determining an attention value of the search resource based on the similarity weight; Acquire all the search resources whose attention values ​​are higher than a set value as resources to be searched; Obtaining the similarity between each of the to-be-searched resources and the preference feature vector; The resources to be detected are sorted in order from high to low according to the similarity and then output.

4. The information indexing method for multi-source heterogeneous university databases as claimed in claim 3 is characterized in that: The obtaining of the similarity between each of the resources to be searched and the preference feature vector, and sorting the resources to be detected in descending order according to the similarity, and then outputting the same, comprises: Obtaining the matching degree between each of the to-be-searched resources and the target search information; Arrange the resources to be searched in descending order according to the matching degree; Selecting the top ranked resources to be searched, and obtaining the similarity between the resources and the preference feature vector; According to the similarity from high to low, the resources to be searched that are ranked at the top are ranked in sequence and then output.

5. The information indexing method for multi-source heterogeneous university databases as claimed in claim 4 is characterized in that: The matching degree is obtained based on the TFIDF algorithm.

6. The information indexing method for multi-source heterogeneous university databases as claimed in claim 1 is characterized in that: Determining a target candidate item in the metadata vector set based on the frequent word vector set includes: Based on the frequent word vector set, obtaining all resources associated with the plurality of historical search information in the metadata vector set as a candidate item set; Obtaining the distance between each element in the candidate item set and the word vector frequent set as the support of the element; The target candidate item is determined based on the element with the smallest support.

7. The information indexing method for multi-source heterogeneous university databases as claimed in claim 6 is characterized in that: The target candidate items satisfy the set search association rules.

8. The information indexing method for multi-source heterogeneous university databases as claimed in claim 1 is characterized in that: The reconstructing the metadata vector set based on the database comprises: Obtaining metadata in the database; The metadata is standardized, deduplicated and integrated according to the RDF format standard using conversion technology; The processed metadata are converted into word vector form based on the Skipgram word vector calculation method.

9. An electronic device, characterized in that: include: a memory for storing instructions executed by one or more processors of the electronic device; The processor, when the processor executes the instructions in the memory, can enable the electronic device to execute the information indexing method for multi-source heterogeneous university databases as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method according to any one of claims 1 to 8.

Images