Method for processing data, electronic device and computer program product

Abstract

Embodiments of the present disclosure provide a method, electronic device, and computer program product for processing data. The method includes training a classification model based on multiple reference documents describing different objects, wherein the trained classification model associates the multiple reference documents with the described objects. The method also includes determining identification information capable of identifying the object from individual words in the reference documents based on their contribution to the association. Utilizing the data processing scheme of the present disclosure, identification information capable of identifying the object in documents describing the object can be determined, thereby automatically generating an identification information dataset for training a machine learning model for determining the identification information.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of data processing, and in particular, to a method, an electronic device, and a computer program product for determining identification information. Background Art

[0002] With the advent of the big data era, the problem of personal privacy leakage is becoming increasingly serious. Therefore, the importance of detecting and identifying personal privacy data in data and protecting it is becoming increasingly prominent. The concept of personally identifiable information (PII) has been introduced. PII is defined as information that can identify a natural person, such as the natural person's name, ID number, location data, and identifiers specific to the natural person's physical, physiological, genetic, economic, and social factors. Currently, in the field of data processing, methods that can identify, detect, and collect PII are urgently needed. Summary of the Invention

[0003] An embodiment of the present disclosure provides a solution for processing data.

[0004] In a first aspect of the present disclosure, a data processing method is provided. The method includes training a classification model based on multiple reference documents describing different objects, wherein the trained classification model associates the multiple reference documents with the described objects. The method also includes determining identification information capable of identifying the object from individual words in the reference documents based on the contribution of the individual words to the association.

[0005] In a second aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor and a memory coupled to the processor, the memory having instructions stored therein, which, when executed by the processor, cause the device to perform an action. The action includes training a classification model based on multiple reference documents describing different objects, wherein the trained classification model associates the multiple reference documents with the described objects. The action also includes determining identification information capable of identifying the object from individual words in the reference documents based on the contribution of the individual words to the association.

[0006] In a third aspect of the present disclosure, there is provided a computer program product tangibly stored on a computer-readable medium and comprising machine-executable instructions which, when executed, cause a machine to perform the method according to the first aspect.

[0007] In an embodiment of the present disclosure, the data processing solution of the present disclosure can be used to determine identification information that can identify an object in a document describing the object, thereby automatically generating an identification information data set for training a machine learning model for determining the identification information.

[0008] This summary is provided to introduce a selection of concepts in a simplified form that are further described in the detailed description below. This summary is not intended to identify key features or essential features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other objects, features and advantages of the embodiments of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components in the exemplary embodiments of the present disclosure.

[0010] Figure 1 A schematic diagram illustrating an environment in which embodiments of the present disclosure can be implemented;

[0011] Figure 2 A flowchart illustrating a method for determining identification information according to some embodiments of the present disclosure is shown;

[0012] Figure 3 shows a schematic diagram of a training classification model according to some embodiments of the present disclosure;

[0013] Figure 4 A schematic diagram illustrating generating an identification information dataset using a trained classification model according to some embodiments of the present disclosure is shown;

[0014] Figure 5 A schematic diagram illustrating training a named entity recognition model based on an identification information dataset and determining identification information in a target document using the trained named entity recognition model according to some embodiments of the present disclosure is shown; and

[0015] Figure 6 A block diagram is shown of an example computing device that can be used to implement embodiments of the present disclosure. DETAILED DESCRIPTION

[0016] The principles of the embodiments of the present disclosure will be described below with reference to several exemplary embodiments shown in the accompanying drawings. Although preferred embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that these embodiments are described only to enable those skilled in the art to better understand and implement the embodiments of the present disclosure, and are not intended to limit the scope of the present disclosure in any way.

[0017] As used herein, the term "including" and its variations represent open inclusion, i.e., "including but not limited to." Unless otherwise stated, the term "or" means "and / or." The term "based on" means "based at least in part on." The terms "one example embodiment" and "some embodiments" mean "at least one example embodiment." The term "another embodiment" means "at least one additional embodiment." The terms "first," "second," etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0018] As used herein, the term "object" may include "individual," "organization," and the like. As used herein, the term "identification information" may include "personally identifiable information" or "PII." PII may include personal privacy data such as name, ID number, and work experience that can identify a specific individual. PII can be classified into direct PII and indirect PII. Direct PII can be used to directly identify a specific individual. Examples of direct PII may include name, phone number, ID number, and the like. Indirect PII may be related to the activities of a specific individual. For example, examples of indirect PII may include an individual's medical history, criminal record, work experience, and the like. As used herein, the terms "PII determination" and "determining PII" may include "PII identification" and "identifying PII."

[0019] As mentioned above, the identification, collection, storage, and processing of PII are highly desirable capabilities in the data industry. Currently, most PII identification methods are only capable of identifying direct PII in data. For example, direct PII such as ID numbers and phone numbers can be identified in data through pattern matching. This is because ID numbers and phone numbers have specific data patterns, making them easy to identify. However, most indirect PII lacks specific data patterns and is therefore difficult to identify.

[0020] For example, if existing PII identification methods were used to remove PII from AAA BBB's autobiographical document, all direct PII, such as name, location, ID card, date of birth, and email address, would no longer exist. However, many fragments of information not closely related to BBB would still remain in the document. When these fragments are combined, one could still determine that the entire document describes the founder of ZZ Company based on AAA BBB's personal work experience, famous quotes, and product design stories. Therefore, improved methods are needed to identify indirect PII in documents.

[0021] Named Entity Recognition (NER) can partially achieve the identification of direct and indirect PII. NER is an information extraction method in natural language processing (NLP) technology. NER can locate entities in unstructured or semi-structured text. These entities can be a variety of things, such as specific individuals or very specific things. For example, NER can extract text containing PII and label it with specific tags (such as people, addresses, etc.).

[0022] NER for PII identification typically includes ontology-based NER and deep learning NER that leverages PII datasets. Ontology-based NER requires a knowledge base. This knowledge base, called an ontology, contains words, terms, and their relationships. Ontology-based NER is a regular expression-based approach that excels at identifying known terms and concepts in unstructured or semi-structured text with specific patterns, but it relies heavily on updating the ontology.

[0023] Deep learning NER is more accurate than the ontology-based NER described above. This is because deep learning NER can cluster words using word embedding technology. This technology clusters words with similar semantic and grammatical relationships. Furthermore, deep learning NER can identify terms and concepts that do not appear in the ontology, giving it strong generalization capabilities.

[0024] However, deep learning NER is a supervised machine learning algorithm that requires a training dataset. This training dataset must include identified, labeled, or annotated PII. Currently, training datasets for deep learning NER for PII recognition include general datasets such as CoNLL 2003, OntoNotes5.0 / CoNLL 2012, and I2b2 Challenges. In these datasets, certain entities or words in documents are labeled with specific tags. Examples of tags include: name, organization, geographic location, activity, product, time indication, etc. As can be seen, although general datasets can provide labeled text content, they only include a limited number of tag categories, and many of these tags are unrelated to PII. For example, tags such as product and time indication may not be related to PII. Therefore, deep learning NER trained on general datasets can only recognize a limited number of predefined PII and has difficulty identifying undefined PII, especially indirect PII.

[0025] According to embodiments of the present disclosure, a data processing solution is proposed to address at least one or more of the aforementioned issues or other potential issues. In this solution, a classification model is trained based on multiple reference documents describing different objects. The trained classification model associates each of the reference documents with the described objects. Then, based on the contribution of individual words in the reference documents to the association, identification information capable of identifying the object is determined from the individual words.

[0026] In this way, the solution can determine identification information in the reference document, thereby automatically generating an identification information dataset without the need for manual labeling of identification information tags. Therefore, a significant amount of time and resources can be saved. In addition, in this solution, by determining specific words in the document as identification information based on the contribution of the words in the document to the association of the document to the corresponding object, potential identification information can be determined. For example, when the object is an individual, the solution can determine potential indirect PII based on the context. This is because when manually labeling PII tags, some words or sentences that can indirectly identify individuals may be missed. Therefore, deep learning NER trained on a general dataset of manually labeled PII may not be able to identify these types of indirect PII.

[0027] In contrast, in this solution, by training a classification model that associates reference documents with corresponding individuals, the contribution of each word in the document to identifying the individual can be determined based on the trained classification model. By identifying words that contribute significantly to the association between the document and the individual as PII, as much potential indirect PII as possible can be determined. Furthermore, by adjusting the contribution threshold, PII datasets of varying quality can be generated. For example, if the contribution threshold is high, only words that contribute significantly to the association can be identified as PII, resulting in a higher-quality PII dataset.

[0028] The following references Figures 1 to 5 It should be understood that these exemplary embodiments are provided only to enable those skilled in the art to better understand and implement the embodiments of the present disclosure, and are not intended to limit the scope of the present disclosure in any way.

[0029] Figure 1 1 shows an example environment 100 in which embodiments of the present disclosure can be implemented. Figure 1As shown, the environment 100 includes a reference data set 110, a computing device 120, and an identification information data set 150. A classification model 135 is deployed in the computing device 120. The computing device 120 includes a computing device 120 in the form of a general-purpose computing device. In some implementations, the computing device 120 can be implemented as various user terminals or service terminals with computing capabilities. The service terminal can be a server, a large computing device, etc. provided by various service providers. The user terminal is such as any type of mobile terminal, fixed terminal, or portable terminal, including a mobile phone, a station, a unit, a device, a multimedia computer, a multimedia tablet, an Internet node, a communicator, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a personal communication system (PCS) device, a personal navigation device, a personal digital assistant (PDA), an audio / video player, a digital camera / camcorder, a positioning device, a television receiver, a radio broadcast receiver, an e-book device, a gaming device, or any combination thereof, including accessories and peripherals of these devices, or any combination thereof.

[0030] The components of computing device 120 may include, but are not limited to, one or more processors or processing units, memory, storage devices, one or more communication units, one or more input devices, and one or more output devices. These components may be integrated into a single device or arranged in a cloud computing architecture. In a cloud computing architecture, these components may be remotely located and work together to implement the functionality described herein. In some implementations, cloud computing provides computing, software, data access, and storage services that do not require the end user to be aware of the physical location or configuration of the systems or hardware providing these services. In various implementations, cloud computing provides services over a wide area network (such as the Internet) using appropriate protocols. For example, a cloud computing provider provides applications over a wide area network, and these applications can be accessed through a web browser or any other computing component. The software or components of the cloud computing architecture and the corresponding data may be stored on servers at remote locations. Computing resources in a cloud computing environment may be consolidated at remote data center locations or they may be distributed. Cloud computing infrastructure may provide services through shared data centers, even though they appear to be a single access point for users. Thus, the components and functionality described herein may be provided by a service provider at a remote location using a cloud computing architecture. Alternatively, they may be provided from conventional servers, or they may be installed directly or otherwise on the client device.

[0031] The computing device 120 may be used to implement the data processing method according to the embodiment of the present disclosure. Figure 1As shown, the computing device 120 receives a reference data set 110. The computing device 120 can receive the reference data set 110 from another computing device or storage device through its input device. The reference data set 110 can have multiple pairs consisting of reference documents 112 describing an object and corresponding object identifiers 114 of the described object. In the case where the object is an individual, examples of the reference documents 112 describing the individual may include biographies, medical records, resumes, bills, criminal records, etc. The object identifier 114 may be a personal identifier. The personal identifier may include the individual's name, ID number, email address, etc. For example, the reference document 112 may be a biography of AAA·BBB, the founder of ZZ Company, and the corresponding personal identifier is the name AAA·BBB. In the case where the object is an organization, examples of the reference documents 112 describing the object may include a brief history of the company, a school history, etc. Examples of the corresponding object identifier 114 may include the organization name, organization code, etc.

[0032] The computing device 120 can use the reference dataset 110 to train a classification model 135 deployed in the computing device 120. The classification model 135 can be a machine learning model configured to associate documents describing an object with the described object. For example, the classification model 135 can be a hierarchical attention network (HAN). The HAN is a commonly used machine learning model for text classification that uses an attention mechanism. After being trained, the classification model 135 can be used to generate an identification information dataset 150 based on the reference documents 112 in the reference dataset 110. This aspect will be referred to below. Figure 2-4 To describe in detail.

[0033] The generated identification information dataset 150 may be composed of the reference documents 112 including the determined identification information. For example, an example of the reference document 112 including the determined identification information is as follows: Figure 1 Document 160 is shown. Document 160 is a description of AAA BBB. Document 160 includes multiple words identified as PII, shown in gray, such as "February 24, 1995," "AAA BBB," "San Francisco, California, USA," and "computer." It should be understood that whether a word is identified as PII depends on the context. For example, in the description of AAA BBB, "computer" might be identified as PII, whereas in a document describing EEE, even if "computer" appears, it might not be identified as PII.

[0034] It should be understood that Figure 1The illustrated environment 100 is merely exemplary and should not be construed as limiting the functionality and scope of the implementations described herein. For example, the computing device 120 may also receive the reference dataset 110 from a storage device integrated therewith. The computing device 120 may also store the generated PII dataset 150 in a storage device integrated therewith.

[0035] Figure 2 1 is a flow chart showing an example method 200 for data processing according to an embodiment of the present disclosure. The method 200 may be performed by, for example, Figure 1 It should be understood that the method 200 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present disclosure is not limited in this respect. Figure 1 The method 200 will be described in detail.

[0036] At block 210, a classification model 135 is trained based on multiple reference documents 112 describing different objects. The trained classification model associates each of the multiple reference documents 112 with the described objects. In some embodiments, the reference documents 112 may include at least one of a biography, a medical record, a resume, a bill, or a criminal record. For example, the reference document 112 may be a biography describing AAA·BBB. In some embodiments, multiple reference documents 112 describing different individuals may be obtained from a public encyclopedia biographical dataset for use in training the classification model 135. In some embodiments, the multiple reference documents 112 describing different individuals have respective personal identifiers of the described individuals. In this case, associating each of the multiple reference documents 112 with the described objects may include associating the multiple reference documents 112 with corresponding personal identifiers. Examples of personal identifiers may include a name, a phone number, an email address, etc. For example, a reference document 112 describing AAA·BBB may have a corresponding personal identifier "AAA·BBB."

[0037] The classification model 135 is trained to associate the plurality of reference documents 112 with the described objects, respectively. In some embodiments, associating the plurality of reference documents 112 with the described objects, respectively, may include associating the plurality of reference documents 112 with the corresponding object identifiers 114, respectively. As described above, the classification model 135 may be a HAN with an attention mechanism. The HAN may be trained based on a plurality of pairs consisting of reference documents 112 and corresponding object identifiers 114 as input data sets. The trained HAN may associate the reference documents 112 with their corresponding object identifiers 114. For example, the trained HAN may associate the reference document 112 describing AAA·BBB with AAA·BBB instead of MM·NN. Details of training the classification model 135 will be described below with reference to Figure 3 Provide a description.

[0038] Figure 3 FIG. 300 shows a schematic diagram of training a classification model 135 according to some embodiments of the present disclosure. Referring to Figure 3 The described process can be regarded as a specific implementation of block 210. As Figure 3 shown, a plurality of reference documents 112 describing different objects and their respective object identifiers 114 can be used to train the classification model 135. When training the classification model 135, the input data can be multiple key-value pairs of the reference document 112 and the corresponding object identifier 114. Conventional training methods can be used to train the classification model 135, that is, by iterative training, the difference between the object identifier 330 predicted by the classification model 135 and the object identifier 114 in the input data is minimized, so as to determine the model parameters that associate the reference document 112 with its corresponding object identifier 114.

[0039] The classification model 135 can be a classification model with an attention mechanism, such as HAN. When training, the classification model 135 with an attention mechanism can update the attention weights assigned to individual words in the document, so as to learn which words in the document contribute more to the association. The scope of the present disclosure is not limited in terms of the specific architecture and training details of the classification model 135, and any classification model 135 that can associate a reference document with a corresponding object and learn the contribution of words in the document to the association can be used.

[0040] Continuing to refer to Figure 2 , at block 220, based on the contribution of individual words in the reference document 112 to the association, identification information that can identify the object is determined from the individual words. The individual words can be each character in the reference document 112, or words after word segmentation. For example, examples of individual words can include "of", "work", "AAA·BBB", etc. As described above, after being trained, the classification model 135 associates a plurality of reference documents 112 with the corresponding described objects. Therefore, the classification model 135 learns through training the contribution of each word in the reference document 112 to associating the reference document 112 with the corresponding object. In other words, the classification model 135 learns through training which words in the reference document 112 are more helpful for identifying the corresponding object. For example, after training is completed, the classification model 135 can learn that the word "ZZ" has a greater contribution to associating the reference document 112 describing AAA·BBB with AAA·BBB, while the word "work" has a smaller contribution to the association. In this case, the word "ZZ" can be determined as the PII for this reference document 112, while the word "work" may not be determined as the PII. Details of determining the PII in the reference document 112 based on the contribution of words to the association will be described below referring to Figure 4 will be described.

[0041] Figure 4 A schematic diagram 400 is shown for generating an identification information dataset 150 using a trained classification model 410 according to some embodiments of the present disclosure. Figure 4 A trained classification model 410 is shown. The trained classification model 410 refers to the classification model 135 after training is completed. Training completion may mean that the classification effect of the trained classification model has exceeded a predetermined value. For example, if the classification accuracy of the classification model 135 reaches 99.99%, the training of the classification model 135 can be considered complete. In some embodiments, after training is completed, the weights assigned to individual words in the reference document 112 by the trained classification model 410 can be determined, and words with weights exceeding a threshold are determined as identification information. The weights can be attention weights that indicate the contribution of the words to the association. For example, the most recently updated attention weights can be extracted from the trained classification model 410, and the words corresponding to the attention weights exceeding the threshold are determined as identification information. It should be understood that the trained classification model 410 can store the attention weights for each word in each of the multiple reference documents 112 used as input. Therefore, after training is completed, the identification information in a specific reference document 112 can be directly determined based on the attention weights assigned to each word in the model.

[0042] Alternatively, in some embodiments, consider the case where the classification model 135 is trained using batches of reference documents 112. Since only a portion of the reference documents 112 is used to train the classification model 135 each time, the trained classification model 410 can only store weights for individual words in a portion of the reference documents 112 used as input. Therefore, only a portion of the reference documents 112 that include the determined identification information may be obtained. In this case, the reference documents 112 in the previous batch of training data sets can be input into the trained classification model 410 again. When the trained classification model 410 associates these reference documents 112 with the described objects, the individual words in the reference documents 112 are assigned attention weights. Then, the words whose attention weights exceed the threshold can be determined as identification information.

[0043] On this basis, the reference documents 112 including the determined identification information can be combined into an identification information dataset 150. For example, the identification information dataset 150 can be composed of multiple Figure 1 The document 160 shown is composed of Figure 1As shown, document 160 includes information on whether each word is determined to be identification information. Words determined to be identification information are shown with a gray background. In some embodiments, a portion of reference documents 112 that include the determined identification information can be selected to form identification information dataset 150. For example, reference documents 112 that include more identification information can be selected to form identification information dataset 150.

[0044] In this way, the identification information dataset 150 can be automatically generated by extracting the learned attention information from the trained classification model 410. This method of automatically generating the identification information dataset 150 can save a lot of manpower and time resources. For example, by extracting attention weights indicating the contribution of words in the reference document 112 to the association from the trained classification model 410, and determining specific words with weights greater than a threshold as identification information capable of identifying the object, the identification information dataset 150 can be automatically generated. Therefore, in an embodiment of the present disclosure, the identification information dataset 150 can be automatically generated using public text such as an encyclopedia biography dataset without the need for manual annotation. In addition, as described above, the identification information is determined based on the contribution of the words to the association, so indirect PII can also be identified and not missed. This can solve the problem of omissions when manually annotating PII.

[0045] The generated identification information dataset 150 can be used to train a NER model so that the trained NER model can determine the identification information in the target document. Figure 5 The process of training the NER model using the generated identification information dataset 150 and determining the identification information in the target document using the trained NER model is described in detail.

[0046] Figure 5 A schematic diagram 500 is shown of training a NER model using the identification information dataset 150 and determining PII in a target document using the trained NER model according to some embodiments of the present disclosure. Figure 5 An identification information dataset 150 , a NER model 510 , a target document 530 , and a trained NER model 540 are shown.

[0047] First, the training phase 501 of training the NER model 510 using the identification information dataset 150 is described. The NER model 510 can be the deep learning NER model described above for determining identification information in a document. For example, the NER model 510 can be a long short-term memory model (CRF+LSTM) using conditional random fields. The process of training the NER model 510 is similar to the process of training the classification model 135, because both the NER model 510 and the classification model 135 are supervised machine learning models. For example, the NER model 510 can be trained based on the identification information dataset 150. Figure 4The generated identification information dataset 150 is used as input for training the NER model 510. The identification information dataset includes a plurality of key-value pairs consisting of the reference document 112 and the identification information determined in the reference document 112. The NER model 510 can be trained by iterative learning to minimize the difference between the identification information 520 predicted by the NER model 510 and the identification information determined in the identification information dataset 150. After the training is completed, the NER model 510 is output 505 as a trained NER model 540.

[0048] In some embodiments, the generated identification information dataset 150 can also be combined with a general dataset for training the NER model 510. Documents in the general dataset may include at least one word manually determined to be identification information. For example, the general dataset includes documents that have been manually annotated with identification information tags. General datasets may include CoNLL 2003, OntoNotes 5.0 / CoNLL 2012, I2b2 Challenges, and the like. In the case of individuals, some of the tags in the general dataset can be reclassified as PII or non-PII, or as having a probability of being PII. This is because the words in the generated identification information dataset 150 can be determined as PII or non-PII, or as having a probability of being PII. By utilizing the generated identification information dataset 150 to augment the general dataset for training the NER model, the need for manually annotated general datasets can be reduced. Furthermore, the generated identification information dataset 150 includes information on both direct and indirect PII, thereby improving the trained NER model's ability to determine indirect PII in target documents. The scope of this disclosure is not limited by the architecture and training details of the NER model 510.

[0049] Continue to refer Figure 5The following describes the inference stage 502 for determining identification information in a target document 530 using a trained NER model 540. For an individual, examples of target document 530 may include medical records, resumes, and the like. Another example is a diary posted online by a user describing their activities on their birthday. After training of NER model 510 is complete, the trained NER model can be used to determine PII in target document 530. Similarly, training completion can mean that the performance of NER model 510 on the training dataset exceeds a threshold. After training is complete, target document 530 can be input into trained NER model 540. Based on the knowledge about PII learned during training, trained NER model 540 can determine PII in target document 530. For example, trained NER model 540 may determine the birthday date and the user's activity history recorded in the diary as PII, while determining the user's birthday wishes recorded in the diary as non-PII.

[0050] In this manner, the NER model 540 trained using the identification information dataset 150 according to an embodiment of the present disclosure can determine the identification information in the target document 130. In the case where the subject is an individual, since the NER model 540 is trained using the identification information dataset 150 containing both direct PII and indirect PII, the trained NER model 540 can determine the indirect PII in the target document 130. Therefore, the trained NER model 540 according to an embodiment of the present disclosure can fully determine the direct PII and indirect PII in the target document 130, thereby reducing the risk of the target document 130 being re-identified even after the determined PII is removed.

[0051] Figure 6 1 shows a schematic block diagram of an example device 600 that can be used to implement embodiments of the present disclosure. Figure 1 The computing device 120 shown may be implemented by the device 600. Figure 6 As shown, the device 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 602 or computer program instructions loaded from a storage unit 608 into a random access memory (RAM) 603. Various programs and data required for the operation of the device 600 can also be stored in the RAM 603. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0052] Various components in device 600 are connected to I / O interface 605, including an input unit 606, such as a keyboard, mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a magnetic disk, optical disk, etc.; and a communication unit 609, such as a network card, modem, wireless communication transceiver, etc. The communication unit 609 allows device 600 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0053] The various processes and processing described above, such as method 200, may be performed by processing unit 601. For example, in some embodiments, method 200 may be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed onto device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by CPU 601, one or more actions of method 200 described above may be performed.

[0054] The present disclosure may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present disclosure.

[0055] Computer-readable storage media can be a tangible device that can hold and store instructions used by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. The computer-readable storage media used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by a fiber optic cable), or an electrical signal transmitted by a wire.

[0056] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0057] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., utilizing an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions. The electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0058] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0059] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine such that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0060] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0061] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and the part of the module, program segment or instruction contains one or more executable instructions for realizing the prescribed logical function. In some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the prescribed function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0062] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for processing data, comprising: A classification model is trained based on a plurality of reference documents describing different objects and a plurality of object identifiers associated with the different objects to obtain a trained classification model, wherein the object identifier includes at least one item selected from the group consisting of: an organization name, an organization code, and a personal identifier, wherein the trained classification model associates the plurality of reference documents with the described objects, respectively, wherein the plurality of reference documents includes at least one item: Biographies, medical records, criminal records, business history information, resumes, and bills; as well as Based on the contribution of individual words in the multiple reference documents to the association, identification information capable of identifying the object is determined from the individual words, wherein the identification information is used to determine personally identifiable information related to at least one of the described objects, wherein the identification information is also used to automatically generate an identification information dataset without manually labeling the identification information, wherein the identification information dataset includes at least a portion of the multiple reference documents.

2. The method according to claim 1, wherein determining identification information capable of identifying the object from the individual words comprises: determining a weight assigned to the individual word by the trained classification model, the weight indicating the contribution; as well as The words whose weights exceed a threshold value among the individual words are determined as the identification information.

3. The method according to claim 1, further comprising: The multiple reference documents including the identification information are combined into the identification information dataset, and the identification information dataset is used to train a named entity recognition (NER) model, so that the trained NER model can determine the identification information in the target document.

4. The method according to claim 3, wherein the identification information dataset is combined with a general dataset to train the NER model, and wherein the documents in the general dataset include at least one word that is manually determined to be identification information.

5. The method according to claim 3 or 4, wherein the NER model is a long short-term memory model using conditional random fields.

6. The method of claim 1, wherein the classification model is a hierarchical attention network.

7. An electronic device comprising: processor; as well as a memory coupled to the processor, the memory having instructions stored therein, the instructions, when executed by the processor, causing the device to perform actions, the actions comprising: A classification model is trained based on a plurality of reference documents describing different objects and a plurality of object identifiers associated with the different objects to obtain a trained classification model, wherein the object identifier includes at least one item selected from the group consisting of: an organization name, an organization code, and a personal identifier, wherein the trained classification model associates the plurality of reference documents with the described objects, respectively, wherein the plurality of reference documents includes at least one item: Biographies, medical records, criminal records, business history information, resumes, and bills; as well as Based on the contribution of individual words in the multiple reference documents to the association, identification information capable of identifying the object is determined from the individual words, wherein the identification information is used to determine personally identifiable information related to at least one of the described objects, wherein the identification information is also used to automatically generate an identification information dataset without manually labeling the identification information, wherein the identification information dataset includes at least a portion of the multiple reference documents.

8. The apparatus according to claim 7, wherein determining identification information capable of identifying the object from the individual words comprises: determining a weight assigned to the individual word by the trained classification model, the weight indicating the contribution; as well as The words whose weights exceed a threshold value among the individual words are determined as the identification information.

9. The apparatus of claim 7, wherein the actions further comprise: The plurality of reference documents including the identification information are combined into the identification information dataset, and the identification information dataset is used to train a named entity recognition (NER) model, so that the trained NER model can determine the identification information in the target document.

10. The apparatus of claim 9, wherein the identification information dataset is combined with a general dataset for training the NER model, and wherein documents in the general dataset include at least one word manually determined to be identification information.

11. The apparatus according to claim 9 or 10, wherein the NER model is a long short-term memory model using conditional random fields.

12. The apparatus of claim 7, wherein the classification model is a hierarchical attention network.

13. A computer program product tangibly stored on a computer-readable medium and comprising machine-executable instructions which, when executed, cause a machine to perform the method according to any one of claims 1 to 6.

Images