WO2020253050A1 - 词向量的配置方法、装置、存储介质、电子装置 - Google Patents

词向量的配置方法、装置、存储介质、电子装置 Download PDF

Info

Publication number
WO2020253050A1
WO2020253050A1 PCT/CN2019/117725 CN2019117725W WO2020253050A1 WO 2020253050 A1 WO2020253050 A1 WO 2020253050A1 CN 2019117725 W CN2019117725 W CN 2019117725W WO 2020253050 A1 WO2020253050 A1 WO 2020253050A1
Authority
WO
WIPO (PCT)
Prior art keywords
vocabulary
word
sequence
stroke
word vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/117725
Other languages
English (en)
French (fr)
Inventor
郑立颖
徐亮
阮晓雯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020253050A1 publication Critical patent/WO2020253050A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • This application relates to the field of neural networks, and specifically to a method, device, storage medium, and electronic device for configuring word vectors.
  • the most basic step is usually to segment words and train word vectors (for example, using the word2vec method for training), and then perform subsequent text comparison and classification tasks based on the word vectors.
  • the inventor found that in the actual processing process, it often happens that the text to be processed contains new words (unregistered words) that are not within the scope of the word vector dictionary.
  • the usual processing method is to randomly assign word vectors to unregistered words, but , The randomly assigned word vectors did not use the semantic information of the new words, resulting in a decrease in the accuracy of subsequent tasks.
  • the embodiments of the application provide a word vector configuration method, device, storage medium, and electronic device, to at least solve the technology that causes the accuracy of subsequent training tasks to decrease when the word vector of an unregistered word is configured by random allocation in the prior art problem.
  • a word vector configuration method including: determining the first word of the initial word vector to be configured; determining whether the first word is in the word vector dictionary, wherein the word vector dictionary is used for storing One-to-one correspondence between multiple words and multiple word vectors; if it is determined that the first word is not in the word vector dictionary, perform stroke disassembly on the first word to obtain the stroke sequence; calculate the strokes of each word in the word vector dictionary The similarity between the sequence and the stroke sequence of the first vocabulary; determine the word vector corresponding to the vocabulary with the highest similarity to the stroke sequence of the first vocabulary, and configure it as the initial word vector of the first vocabulary.
  • the word vector configuration method provided in this application, by disassembling the vocabulary that is not registered in the word vector dictionary, find the registered word closest to its stroke in the word vector dictionary, and then configure the word vector of words with similar strokes It is the initial word vector of the unregistered word, which solves the technical problem that the accuracy of subsequent training tasks decreases when the word vector of the unregistered word is configured by random allocation in the related technology.
  • the semantic information carried in the Chinese strokes is used to identify the unregistered word. Assigning initial word vectors to login words can reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • a word vector configuration device including: a first determination module, used to determine the first word of the initial word vector to be configured; a judgment module, used to determine whether the first word is in In the word vector dictionary, the word vector dictionary is used to store the one-to-one correspondence between multiple words and multiple word vectors; the disassembly module is used to execute the first word if it is determined that the first word is not in the word vector dictionary The strokes are disassembled to obtain the stroke sequence; the calculation module is used to calculate the similarity between the stroke sequence of each word in the word vector dictionary and the stroke sequence of the first word; the second determination module is used to determine the stroke sequence of the first word The word vector corresponding to the word with the highest sequence similarity is configured as the initial word vector of the first word.
  • the word vector dictionary is searched for the registered word closest to its stroke, and then the word vectors of the words with similar strokes are arranged It is the initial word vector of the unregistered word, which solves the technical problem that the accuracy of subsequent training tasks decreases when the word vector of the unregistered word is configured by random allocation in the related technology.
  • the semantic information carried in the Chinese strokes is used to identify the unregistered word. Assigning initial word vectors to login words can reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • non-volatile readable storage medium in which computer readable instructions are stored, wherein the computer readable instructions are configured to execute the above Steps in any method embodiment.
  • an electronic device including a memory and a processor, the memory stores computer-readable instructions, and the processor is configured to run the computer-readable instructions to execute Steps in any of the above method embodiments.
  • Fig. 1 is a flowchart of a word vector configuration method according to an embodiment of the present application
  • Fig. 2 is a schematic diagram of a word vector configuration device according to an embodiment of the present application.
  • Fig. 3 is a block diagram of the hardware structure of an electronic device according to an embodiment of the present application.
  • This embodiment provides a method for configuring word vectors, which can be applied to mobile terminals, handheld terminals or similar computing devices. Running on different computing devices is only the difference in the implementation of the scheme. Those skilled in the art can foresee that running on different computing devices can produce the same technical effect.
  • the word vector configuration method provided in this embodiment disassembles the vocabulary that is not registered in the word vector dictionary, finds the registered word closest to its stroke in the word vector dictionary, and then configures the word vectors of words with similar strokes It is the initial word vector of the unregistered word, which solves the technical problem that the accuracy of subsequent training tasks decreases when the word vector of the unregistered word is configured by random allocation in the related technology.
  • the semantic information carried in the Chinese strokes is used to identify the unregistered word. Assigning initial word vectors to login words can reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • the word vector configuration method provided in this embodiment includes the following steps:
  • Step 101 Determine the first vocabulary of the initial word vector to be configured.
  • the first vocabulary can be any participle vocabulary in the corpus to be processed.
  • This embodiment is applied to configure the word vectors of the vocabulary. Since the machine is not easy to recognize natural language in the process of neural network training, it is necessary to convert the natural language into machine language so that the machine can recognize different words. In this embodiment It uses word vectors to identify different words.
  • the neural network training may be realized through a neural network model, which includes but is not limited to a BP neural network model, a CNN neural network model, and the like.
  • determining the vocabulary based on the corpus first perform word segmentation processing on the expected, and then determine that the vocabulary for which the initial word vector needs to be configured is the first vocabulary. Specifically, determining the first vocabulary of the initial word vector to be configured includes the following steps:
  • Step 11 Obtain the corpus of the word to be segmented
  • Step 12 Perform word segmentation on the corpus to obtain multiple word segments in sequence;
  • Step 13 Determine the first participle for which no initial word vector is configured among the multiple participles to obtain the first vocabulary.
  • the word segmentation processing can use the existing word segmentation algorithm, which will not be repeated in this embodiment.
  • Step 102 Determine whether the first vocabulary is in the word vector dictionary, where the word vector dictionary is used to store a one-to-one correspondence between multiple words and multiple word vectors.
  • the word vector dictionary is used to store the relational pairs of vocabulary and word vector that have been trained or manually labeled. After obtaining the first vocabulary, it is determined whether the first vocabulary is in the word vector dictionary. If it is determined that the first vocabulary is in the word vector dictionary, the word vector corresponding to the first vocabulary is searched in the word vector dictionary.
  • Step 103 If it is judged that the first vocabulary is not in the word vector dictionary, perform stroke disassembly on the first vocabulary to obtain the stroke sequence; when performing stroke disassembly on the first vocabulary, each word in the vocabulary may be disassembled into
  • the strokes of the smallest unit of Chinese, such as horizontal, vertical, skew, ⁇ , etc. can also be used to disassemble each word in the vocabulary to a preset level according to the preset smallest unit, for example, the sun, moon, wood, Commonly used constituent units such as head, mouth, field, etc. are used as a minimum unit to reduce the length of the stroke sequence obtained after the simple strokes are disassembled.
  • the specific degree of disassembly can be determined according to actual settings, which is not specifically limited in this embodiment, and will not be repeated here.
  • the stroke sequence is the sequence of all strokes included in the corresponding vocabulary. Different strokes can be identified by a combination of characters such as numbers and letters.
  • the sequence of identifiers of all strokes is the stroke sequence. For example, if the identifiers of the horizontal and vertical strokes are 1, 2, 3, and 4, the stroke sequence of the word "Wood" is ⁇ 1, 2, 3, 4 ⁇ . For another example, if the identifier of "Wood" is 1, and the destination identifier is 2, then the stroke sequence of the word "Phase" is ⁇ 1, 2 ⁇ .
  • Step 104 Calculate the similarity between the stroke sequence of each vocabulary in the word vector dictionary and the stroke sequence of the first vocabulary; this step may be executed every time the stroke sequence of a vocabulary is obtained through step 103, that is, every time After disassembling a word in a corpus, calculate its similarity with each word in the word vector dictionary. Or, it can also be executed after disassembling all unregistered words (words not in the word vector dictionary) in a certain corpus. In addition, when calculating the similarity of the stroke sequence between a vocabulary and each vocabulary in the word vector dictionary, it can also be executed synchronously or asynchronously. The synchronous execution is divided into several processes, and each process calculates the vocabulary and the word vector dictionary The similarity of a vocabulary in the Asynchronous execution means that the similarity between the vocabulary and each vocabulary in the word vector dictionary is calculated in sequence through only one process.
  • Step 21 Determine the total length of the overlapping sequence segment of the stroke sequence of the second vocabulary and the stroke sequence of the first vocabulary, where the overlapping sequence segment is a sequence segment with the same stroke arrangement in the stroke sequences of the two vocabulary.
  • the stroke sequence of the first word is ⁇ 5,3,4,9,7,1,3,13 ⁇
  • the stroke sequence of the second word is ⁇ 1,3,10,5,3,4,9,11 ⁇ .
  • there are two overlapping sequence segments of the first vocabulary and the second vocabulary which are 1, 3 and 5, 3, 4, 9, and the length is 2 and 4 respectively.
  • Step 22 Determine the similarity between the stroke sequence of the first vocabulary and the stroke sequence of the second vocabulary based on the total length of the overlapping sequence segments of the first vocabulary and the second vocabulary.
  • the following formula may be used:
  • step 103 is described in detail as follows:
  • Step 1 Disassemble the new words according to the strokes to form a sequence of strokes
  • Step 2 Disassemble all the words in the existing word vector model according to the strokes, and get a stroke sequence for each word;
  • Step 3 Compare the sequence value obtained in Step 1 with all the sequence sets in Step 2 to find the most similar sequence.
  • the similarity calculation method is: 2*(length of coexisting sequence)/(length of sequence 1+ Length of sequence 2).
  • sequence 1 is abcd
  • sequence 2 is bcde
  • the length of the co-existing sequence is 3 (bcd)
  • Step 4 Find the most similar sequence and assign the word vector corresponding to the sequence to the new word.
  • Step 105 Determine the word vector corresponding to the vocabulary with the highest degree of similarity in the stroke sequence of the first vocabulary, and configure it as the initial word vector of the first vocabulary. That is, if it is determined that the first word is not in the word vector dictionary, compare the strokes of the first word with the strokes of the words in the word vector dictionary, find the word with the highest stroke sequence similarity, and compare the strokes in the word vector dictionary The word vector of the word with the highest sequence similarity is assigned to the first word.
  • This technical solution can make full use of the rich word vector dictionary, use the stroke composition information of Chinese itself to expand its semantic features, use the word vector of the known vocabulary to predict the word vector of the unknown vocabulary, and make up for the defect of random assignment of unregistered words in the word vector , And improve the accuracy of subsequent training tasks and reduce the training gradient.
  • the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical sequence is shown in the flowchart, in some cases, The steps shown or described can be performed in a different order than here.
  • the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is Better implementation.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the method described in each embodiment of the present application.
  • a device for configuring word vectors is also provided, which is used to implement the above-mentioned embodiment 1 and its preferred embodiments.
  • a device for configuring word vectors is also provided, which is used to implement the above-mentioned embodiment 1 and its preferred embodiments.
  • module is a combination of software and/or hardware that can implement predetermined functions.
  • the devices described in the following embodiments are preferably implemented by software, the implementation of hardware or a combination of software and hardware can also be conceived.
  • FIG. 2 is a schematic diagram of a word vector configuration device according to an embodiment of the present application. As shown in FIG. 2, the device includes: a first determination module 10, a judgment module 20, a disassembly module 30, a calculation module 40, and a second determination module 50.
  • the first determination module is used to determine the first vocabulary of the initial word vector to be configured; the judgment module is used to determine whether the first vocabulary is in the word vector dictionary, where the word vector dictionary is used to store multiple words and multiple words The one-to-one correspondence of vectors; the disassembly module is used to perform stroke disassembly on the first vocabulary if it is determined that the first word is not in the word vector dictionary to obtain the stroke sequence; the calculation module is used to calculate each word in the word vector dictionary The similarity between the stroke sequence of a vocabulary and the stroke sequence of the first vocabulary; the second determining module is used to determine the word vector corresponding to the vocabulary with the highest similarity to the stroke sequence of the first vocabulary, and configure it as the initial word of the first vocabulary vector.
  • the word vector dictionary includes the second vocabulary
  • the calculation module includes: a first determining unit configured to determine the total length of the overlap sequence segment of the stroke sequence of the second vocabulary and the stroke sequence of the first vocabulary, wherein the overlap sequence A segment is a sequence segment with the same stroke arrangement in the stroke sequences of two words; the second determining unit is used to determine the stroke sequence of the first word and the second word based on the total length of the overlapping sequence of the first word and the second word The similarity of the stroke sequence.
  • the first determination module includes: an acquisition unit for acquiring the corpus of the word to be segmented; a word segmentation unit for segmenting the corpus to obtain multiple word segmentation in sequence; and a third determination unit for acquiring the corpus of the words to be segmented
  • the first participle of the initial word vector is not configured, and the first word is obtained.
  • This embodiment disassembles the vocabulary that is not registered in the word vector dictionary, searches the word vector dictionary for the registered word closest to its stroke, and then configures the word vector of the word with similar strokes as the initial word vector of the unregistered word , It solves the technical problem that the accuracy of subsequent training tasks is reduced when the word vectors of unregistered words are configured by random allocation in related technologies.
  • the semantic information carried in Chinese strokes is used to assign initial word vectors to unregistered words. Reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • each of the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following manner, but not limited to this: the above modules are all located in the same processor; or, the above modules are combined in any combination The forms are located in different processors.
  • modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices.
  • they can be implemented with program codes executable by the computing device, so that they can be stored in the storage device for execution by the computing device, and in some cases, can be executed in a different order than here.
  • the embodiment of the present application also provides a non-volatile readable storage medium, the storage medium stores a computer readable instruction, wherein the computer readable instruction is set to execute any of the above method embodiments when running Steps in.
  • This embodiment disassembles the vocabulary that is not registered in the word vector dictionary, searches the word vector dictionary for the registered word closest to its stroke, and then configures the word vector of the word with similar strokes as the initial word vector of the unregistered word , It solves the technical problem that the accuracy of subsequent training tasks is reduced when the word vectors of unregistered words are configured by random allocation in related technologies.
  • the semantic information carried in Chinese strokes is used to assign initial word vectors to unregistered words. Reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • the foregoing storage medium may include, but is not limited to: U disk, Read-Only Memory (Read-Only Memory, ROM for short), Random Access Memory (Random Access Memory, RAM for short), Various media that can store computer programs, such as mobile hard disks, magnetic disks, or optical disks.
  • An embodiment of the present application also provides an electronic device, including a memory and a processor, the memory stores computer readable instructions, and the processor is configured to run the computer readable instructions to execute any of the above method embodiments.
  • a step of. This embodiment disassembles the vocabulary that is not registered in the word vector dictionary, searches the word vector dictionary for the registered word closest to its stroke, and then configures the word vector of the word with similar strokes as the initial word vector of the unregistered word , It solves the technical problem that the accuracy of subsequent training tasks is reduced when the word vectors of unregistered words are configured by random allocation in related technologies.
  • the semantic information carried in Chinese strokes is used to assign initial word vectors to unregistered words. Reduce the time-consuming of subsequent training tasks and improve the accuracy of training tasks.
  • the aforementioned electronic device may further include a transmission device and an input-output device, wherein the transmission device is connected to the aforementioned processor, and the input-output device is connected to the aforementioned processor.
  • FIG. 3 is a hardware structure block diagram of an electronic device according to an embodiment of the present application.
  • the electronic device may include one or more (only one is shown in FIG. 3) processor 302 (the processor 302 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) And a memory 304 for storing data.
  • the above electronic device may also include a transmission device 306 and an input/output device 308 for communication functions.
  • a transmission device 306 may also include a transmission device 306 and an input/output device 308 for communication functions.
  • the memory 304 may be used to store computer-readable instructions, for example, software programs and modules of application software, such as computer-readable instructions corresponding to the image recognition method in the embodiment of the present application.
  • the processor 302 runs the computer stored in the memory 304
  • the instructions are readable to execute various functional applications and data processing, that is, to implement the above methods.
  • the memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 304 may further include a memory remotely provided with respect to the processor 302, and these remote memories may be connected to the electronic device through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the transmission device 306 is used to receive or send data via a network.
  • the foregoing specific examples of the network may include a wireless network provided by a communication provider of an electronic device.
  • the transmission device 306 includes a network adapter (Network Interface Controller, NIC for short), which can be connected to other network devices through a base station to communicate with the Internet.
  • the transmission device 306 may be a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.
  • RF radio frequency

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Machine Translation (AREA)

Abstract

本申请提供了一种词向量的配置方法、装置、存储介质、电子装置，其中，本申请提供的词向量的配置方法包括：确定待配置初始词向量的第一词汇；判断第一词汇是否在词向量词典中，其中，词向量词典用于存储多个词汇与多个词向量的一一对应关系；如果判断出第一词汇不在词向量词典中，对第一词汇执行笔画拆解，得到笔画序列；计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度；确定与第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为第一词汇的初始词向量。通过本申请，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题。

Description

词向量的配置方法、装置、存储介质、电子装置
本申请要求与2019年6月20日提交中国专利局、申请号为2019105348108、申请名称为“词向量的配置方法、装置、存储介质、电子装置”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及神经网络领域，具体而言，涉及一种词向量的配置方法、装置、存储介质、电子装置。
背景技术
在处理文本类数据时，通常最基本的步骤就是分词并训练词向量(例如，采用word2vec方法进行训练)，然后基于词向量进行后续的文本比对、分类等任务。发明人发现在实际处理过程中，经常会发生待处理的文本中包含不在词向量词典范围内的新词(未登录词)，通常的处理方法是随机对未登录词随机的分配词向量，但是，随机分配的词向量并没有利用到新词的语义信息，造成后续任务精度下降。
针对相关技术中存在的上述问题，目前尚未发现有效的解决方案。
发明内容
本申请实施例提供了一种词向量的配置方法、装置、存储介质、电子装置，以至少解决现有技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题。
根据本申请的一个实施例，提供了一种词向量的配置方法，包括：确定待配置初始词向量的第一词汇；判断第一词汇是否在词向量词典中，其中，词向量词典用于存储多个词汇与多个词向量的一一对应关系；如果判断出第一词汇不在词向量词典中，对第一词汇执行笔画拆解，得到笔画序列；计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度；确定与第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为第一词汇的初始词向量。
通过本申请提供的词向量的配置方法，通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。
根据本申请的另一个实施例，提供了一种词向量的配置装置，包括：第一确定模块，用于确定待配置初始词向量的第一词汇；判断模块，用于判断第一词汇是否在词向量词典中，其中，词向量词典用于存储多个词汇与多个词向量的一一对应关系；拆解模块，用于如果判断出第一词汇 不在词向量词典中，对第一词汇执行笔画拆解，得到笔画序列；计算模块，用于计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度；第二确定模块，用于确定与第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为第一词汇的初始词向量。
通过本申请提供的词向量的配置装置，通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。
根据本申请的又一个实施例，还提供了一种非易失性可读存储介质，所述存储介质中存储有计算机可读指令，其中，所述计算机可读指令被设置为运行时执行上述任一项方法实施例中的步骤。
根据本申请的又一个实施例，还提供了一种电子装置，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器被设置为运行所述计算机可读指令以执行上述任一项方法实施例中的步骤。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例的词向量的配置方法的流程图；
图2是根据本申请实施例的词向量的配置装置的示意图；
图3是本申请实施例的一种电子装置的硬件结构框图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例1
本实施例提供了一种词向量的配置方法，可以应用于移动终端、手持终端或类似的运算设备之中。运行在不同的运算设备仅是方案在执行主体上的差异，本领域人员可预见在不同运算设备中运行能够产生相同的技术效果。本实施例提供的词向量的配置方法，通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。
如图1所示，本实施例提供的词向量的配置方法包括如下步骤：
步骤101，确定待配置初始词向量的第一词汇。第一词汇可以是在待处理的语料中的任意一个分词词汇。本实施例应用于对词汇的词向量进行配置，由于在神经网络训练的过程中，机器不便于识别自然语言，需要将自然语言转变为机器语言，以便于机器识别不同的词汇，本实施例中是通过词向量来标识不同的词汇。在本实施例中，神经网络训练可以是通过神经网络模型实现的，所述神经网络模型包括但不限于BP神经网络模型、CNN神经网络模型等。在基于语料确定词汇时，首先对预料执行分词处理，然后确定当前需要配置初始词向量的词汇为第一词汇。具体而言，确定待配置初始词向量的第一词汇，包括如下步骤：
步骤11，获取待分词的语料；
步骤12，对语料进行分词，得到顺序的多个分词；
步骤13，在多个分词中确定未配置初始词向量的首位分词，得到第一词汇。其中，分词处理可以采用现有的分词算法，本实施例在此不再赘述。
步骤102，判断第一词汇是否在词向量词典中，其中，词向量词典用于存储多个词汇与多个词向量的一一对应关系。词向量词典用于存储已经训练、或已经人工标注过的词汇和词向量的关系对。在获取第一词汇之后，判断第一词汇是否在词向量词典中。如果判断出第一词汇在词向量词典中，则在词向量词典查找第一词汇对应的词向量。
步骤103，如果判断出第一词汇不在词向量词典中，对第一词汇执行笔画拆解，得到笔画序列；在对第一词汇进行笔画拆解时，可以将词汇中的每个字拆解为中文最小单位的笔画，例如，横、竖、撇、捺等等，也可以依据预设的最小单位将词汇中的每个字拆解到预设的程度，例如，将日、月、木、目、口、田等常用的组成单元作为一个最小单位，以化简笔画拆解之后得到的笔画序列的长度。具体拆解到何种程度可以根据实际设定来确定，本实施例对此不作具体限制，在此不再赘述。笔画序列是对应词汇所包括的所有笔画的序列。可以通过数字、字母等字符的组合来标识不同的笔画，所有笔画的标识符组成的序列就是笔画序列。例如，如果横竖撇捺的标识符分别为1、2、3、4，则“木”字的笔画序列为{1，2，3，4}。又如，如果“木”的标识符为1，目的标识符为2，则“相”字的笔画序列为{1，2}。
步骤104，计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度；该步骤可以是在每通过步骤103获取到一个词汇的笔画序列之后执行的，也即，每对一个语料中的一个词汇拆解完毕之后，计算其与词向量词典中每个词汇的相似度。或者，也可以是在对某一段语料中所有未登录词(未在词向量词典中的词)进行拆解完毕后分别执行的。此外，在计算一个词汇与词向量词典中每个词汇的笔画序列相似度时，也可以时同步执行或异步执行的，同步执行即为分若干个进程，每个进程计算该词汇与词向量词典中的一个词汇的相似度，异步执行则为仅通过一个进程，顺序执行计算该词汇与词向量词典中每一个词汇的相似度。
在计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度时，可以采用如下步骤：
步骤21，确定第二词汇的笔画序列与第一词汇的笔画序列的重合序列段的总长度，其中，重合序列段为两个词汇的笔画序列中笔画排列相同的序列段。例如，第一词汇的笔画序列为{5，3，4，9，7，1，3，13}第二词汇的笔画序列为{1，3，10，5，3，4，9，11}，则第一词汇和第二词汇的重合序列段有两段，分别为1，3和5，3，4，9，长度分别为2和4。
步骤22，基于第一词汇与第二词汇的重合序列段的总长度，确定第一词汇的笔画序列与第二词汇的笔画序列的相似度。可选的，在步骤22基于第一词汇与第二词汇的重合序列段的总长度，确定第一词汇的笔画序列与第二词汇的笔画序列的相似度时，可以采用如下公式：
S＝2*p/(n+m)
其中，S为第一词汇的笔画序列与第二词汇的笔画序列的相似度，p为第一词汇与第二词汇的重合序列段的总长度，n为第一词汇的笔画序列的长度，m为第二词汇的笔画序列的长度。举例而言，以一种可选的使用场景为例，对步骤103进行详细说明如下：
步骤一：将新词按笔画进行拆解，形成笔画序列；
步骤二：将已有词向量模型中的所有词，按照笔画进行拆解，每一个词得到一个笔画序列；
步骤三：将步骤一中得到的序列值与步骤二中所有的序列集进行对比，找出最相似的序列，相似计算方法为：2*(共同存在的序列长度)/(序列1的长度+序列2的长度)。例如：序列1为abcd，序列2为bcde，则共同存在的序列长度为3(bcd)，相似度为2*3/(4+4)＝0.75
步骤四：找出最相似的序列后将该序列对应的词的向量赋予给新词。
步骤105，确定与第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为第一词汇的初始词向量。也即，如果判断出第一词汇不在词向量词典中，依据第一词汇的笔画与词向量词典中的词汇的笔画进行比对，寻找笔画序列相似度最高的词汇，并将词向量词典中笔画序列相似度最高的词汇的词向量赋给第一词汇。本技术方案能够充分利用丰富的词向量词典，利用中文本身的笔画构成信息来扩充其语义的特征，利用已知词汇的词向量预测未知词汇的词向量，弥补词向量未登录词随机赋值的缺陷，并提升后续训练任务的精度，减少训练梯度。需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流 程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
实施例2
在本实施例中还提供了一种词向量的配置装置，该装置用于实现上述实施例1及其优选实施方式，对于本实施例中未详述的术语或实现方式，可参见实施例1中的相关说明，已经进行过说明的不再赘述。如以下所使用的术语“模块”，是可以实现预定功能的软件和/或硬件的组合。尽管以下实施例所描述的装置较佳地以软件来实现，但是硬件，或者软件和硬件的组合的实现也是可以被构想的。
图2是根据本申请实施例的词向量的配置装置的示意图，如图2所示，该装置包括：第一确定模块10，判断模块20，拆解模块30，计算模块40和第二确定模块50。其中，第一确定模块用于确定待配置初始词向量的第一词汇；判断模块，用于判断第一词汇是否在词向量词典中，其中，词向量词典用于存储多个词汇与多个词向量的一一对应关系；拆解模块，用于如果判断出第一词汇不在词向量词典中，对第一词汇执行笔画拆解，得到笔画序列；计算模块，用于计算词向量词典中的每个词汇的笔画序列与第一词汇的笔画序列的相似度；第二确定模块，用于确定与第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为第一词汇的初始词向量。
可选的，词向量词典中包括第二词汇，计算模块包括：第一确定单元，用于确定第二词汇的笔画序列与第一词汇的笔画序列的重合序列段的总长度，其中，重合序列段为两个词汇的笔画序列中笔画排列相同的序列段；第二确定单元，用于基于第一词汇与第二词汇的重合序列段的总长度，确定第一词汇的笔画序列与第二词汇的笔画序列的相似度。可选的，第二确定单元基于第一词汇与第二词汇的重合序列段的总长度，确定第一词汇的笔画序列与第二词汇的笔画序列的相似度，采用如下公式：S＝2*p/(n+m)其中，S为第一词汇的笔画序列与第二词汇的笔画序列的相似度，p为第一词汇与第二词汇的重合序列段的总长度，n为第一词汇的笔画序列的长度，m为第二词汇的笔画序列的长度。可选的，第一确定模块包括：获取单元，用于获取待分词的语料；分词单元，用于对语料进行分词，得到顺序的多个分词；第三确定单元，用于在多个分词中确定未配置初始词向量的首位分词，得到第一词汇。
本实施例通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的 笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。需要说明的是，上述各个模块是可以通过软件或硬件来实现的，对于后者，可以通过以下方式实现，但不限于此：上述模块均位于同一处理器中；或者，上述各个模块以任意组合的形式分别位于不同的处理器中。
显然，本领域的技术人员应该明白，上述的本申请的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，可选地，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本申请不限制于任何特定的硬件和软件结合。
实施例3
本申请的实施例还提供了一种非易失性可读存储介质，该存储介质中存储有计算机可读指令，其中，该计算机可读指令被设置为运行时执行上述任一项方法实施例中的步骤。本实施例通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。可选地，在本实施例中，上述存储介质可以包括但不限于：U盘、只读存储器(Read-Only Memory，简称为ROM)、随机存取存储器(Random Access Memory，简称为RAM)、移动硬盘、磁碟或者光盘等各种可以存储计算机程序的介质。
实施例4
本申请的实施例还提供了一种电子装置，包括存储器和处理器，该存储器中存储有计算机可读指令，该处理器被设置为运行计算机可读指令以执行上述任一项方法实施例中的步骤。本实施例通过对未登录在词向量词典中的词汇进行拆解，在词向量词典中查找与其笔画最接近的已登录词，进而将笔画相近词的词向量配置为未登录词的初始词向量，解决了相关技术中利用随机分配的方式配置未登录词的词向量时导致后续训练任务精度下降的技术问题，利用中文的笔画中所携带的语义信息来为未登录词赋予初始词向量，能够减少后续训练任务的耗时、提高训练任务的精度。可选地，上述电子装置还可以包括传输设备以及输入输出设备，其中，该传输设备和上述处理器连接，该输入输出设备和上述处理器连接。以电子装置为电子装置为例，图3是本申请实施例的一种电子装置的硬件结构框图。如图3所示，电子装置可以包括一个或多个(图3中仅示出一个)处理器302(处理器302可以包括但不限于微处理器MCU或可编程逻辑器件FPGA等的处理装置)和用于存储数据的存储器304，可选地，上述电子装置还可以包括用于通信功能的传输设备306以及输入输出设备308。本领域普通技术人员可以理解，图3所示的结构仅为示意，其并不对上述电子装置的结构造成限定。例如，电子装置还可包括比图3中所示更多或者更少的组件， 或者具有与图3所示不同的配置。存储器304可用于存储计算机可读指令，例如，应用软件的软件程序以及模块，如本申请实施例中的图像的识别方法对应的计算机可读指令，处理器302通过运行存储在存储器304内的计算机可读指令，从而执行各种功能应用以及数据处理，即实现上述的方法。存储器304可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器304可进一步包括相对于处理器302远程设置的存储器，这些远程存储器可以通过网络连接至电子装置。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。传输装置306用于经由一个网络接收或者发送数据。上述的网络具体实例可包括电子装置的通信供应商提供的无线网络。在一个实例中，传输装置306包括一个网络适配器(Network Interface Controller，简称为NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输装置306可以为射频(Radio Frequency，简称为RF)模块，其用于通过无线方式与互联网进行通讯。
以上所述仅为本申请的优选实施例而已，并不用于限制本申请，对于本领域的技术人员来说，本申请可以有各种更改和变化。凡在本申请的原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的保护范围之内。

Claims (16)

  1. 一种词向量的配置方法，所述方法包括：
    确定待配置初始词向量的第一词汇；
    判断所述第一词汇是否在词向量词典中，其中，所述词向量词典用于存储多个词汇与多个词向量的一一对应关系；
    如果判断出所述第一词汇不在所述词向量词典中，对所述第一词汇执行笔画拆解，得到笔画序列；
    计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度；
    确定与所述第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为所述第一词汇的初始词向量。
  2. 根据权利要求1所述的方法，所述词向量词典中包括第二词汇，所述计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度，包括：
    确定所述第二词汇的笔画序列与所述第一词汇的笔画序列的重合序列段的总长度，其中，所述重合序列段为两个词汇的笔画序列中笔画排列相同的序列段；
    基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度。
  3. 根据权利要求2所述的方法，所述基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，采用如下公式：
    S＝2*p/(n+m)
    其中，S为所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，p为所述第一词汇与所述第二词汇的重合序列段的总长度，n为所述第一词汇的笔画序列的长度，m为所述第二词汇的笔画序列的长度。
  4. 根据权利要求1所述的方法，所述确定待配置初始词向量的第一词汇，包括：获取待分词的语料；对所述语料进行分词，得到顺序的多个分词；在所述多个分词中确定未配置初始词向量的首位分词，得到所述第一词汇。
  5. 一种词向量的配置装置，包括：
    第一确定模块，用于确定待配置初始词向量的第一词汇；
    判断模块，用于判断所述第一词汇是否在词向量词典中，其中，所述词向量词典用于存储多个词汇与多个词向量的一一对应关系；
    拆解模块，用于如果判断出所述第一词汇不在所述词向量词典中，对所述第一词汇执行笔画拆解，得到笔画序列；
    计算模块，用于计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度；
    第二确定模块，用于确定与所述第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为所述第一词汇的初始词向量。
  6. 根据权利要求5所述的装置，所述词向量词典中包括第二词汇，所述计算模块包括：
    第一确定单元，用于确定所述第二词汇的笔画序列与所述第一词汇的笔画序列的重合序列段的总长度，其中，所述重合序列段为两个词汇的笔画序列中笔画排列相同的序列段；
    第二确定单元，用于基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度。
  7. 根据权利要求6所述的装置，所述第二确定单元基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，采用如下公式：
    S＝2*p/(n+m)
    其中，S为所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，p为所述第一词汇与所述第二词汇的重合序列段的总长度，n为所述第一词汇的笔画序列的长度，m为所述第二词汇的笔画序列的长度。
  8. 根据权利要求5所述的装置，所述第一确定模块包括：
    获取单元，用于获取待分词的语料；
    分词单元，用于对所述语料进行分词，得到顺序的多个分词；
    第三确定单元，用于在所述多个分词中确定未配置初始词向量的首位分词，得到所述第一词汇。
  9. 一种非易失性可读存储介质，所述存储介质中存储有计算机可读指令，其中，所述计算机可读指令被设置为运行时执行词向量的配置方法，包括：确定待配置初始词向量的第一词汇；判断所述第一词汇是否在词向量词典中，其中，所述词向量词典用于存储多个词汇与多个词向量的一一对应关系；如果判断出所述第一词汇不在所述词向量词典中，对所述第一词汇执行笔画拆解，得到笔画序列；计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度；确定与所述第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为所述第一词汇的初始词向量。
  10. 根据权利要求9所述的存储介质，所述词向量词典中包括第二词汇，所述计算机可读指令被设置为运行时执行所述计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度，包括：确定所述第二词汇的笔画序列与所述第一词汇的笔画序列的重合序列段的总长度，其中，所述重合序列段为两个词汇的笔画序列中笔画排列相同的序列段；基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度。
  11. 根据权利要求10所述的存储介质，所述计算机可读指令被设置为运行时执行所述基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，采用如下公式：
    S＝2*p/(n+m)
    其中，S为所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，p为所述第一词汇与所述第二词汇的重合序列段的总长度，n为所述第一词汇的笔画序列的长度，m为所述第二词汇的笔画序列的长度。
  12. 根据权利要求9所述的存储介质，所述计算机可读指令被设置为运行时执行所述确定待配置初始词向量的第一词汇，包括：获取待分词的语料；对所述语料进行分词，得到顺序的多个分词；在所述多个分词中确定未配置初始词向量的首位分词，得到所述第一词汇。
  13. 一种电子装置，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器被设置为运行所述计算机可读指令以执行词向量的配置方法，包括：确定待配置初始词向量的第一词汇；判断所述第一词汇是否在词向量词典中，其中，所述词向量词典用于存储多个词汇与多个词向量的一一对应关系；如果判断出所述第一词汇不在所述词向量词典中，对所述第一词汇执行笔画拆解，得到笔画序列；计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度；确定与所述第一词汇的笔画序列相似度最高的词汇对应的词向量，并配置为所述第一词汇的初始词向量。
  14. 根据权利要求13所述的电子装置，所述词向量词典中包括第二词汇，所述计算机可读指令被设置为运行时执行所述计算所述词向量词典中的每个词汇的笔画序列与所述第一词汇的笔画序列的相似度，包括：确定所述第二词汇的笔画序列与所述第一词汇的笔画序列的重合序列段的总长度，其中，所述重合序列段为两个词汇的笔画序列中笔画排列相同的序列段；基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度。
  15. 根据权利要求14所述的电子装置，所述计算机可读指令被设置为运行时执行所述基于所述第一词汇与所述第二词汇的重合序列段的总长度，确定所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，采用如下公式：
    S＝2*p/(n+m)
    其中，S为所述第一词汇的笔画序列与所述第二词汇的笔画序列的相似度，p为所述第一词汇与所述第二词汇的重合序列段的总长度，n为所述第一词汇的笔画序列的长度，m为所述第二词汇的笔画序列的长度。
  16. 根据权利要求13所述的电子装置，所述计算机可读指令被设置为运行时执行所述确定待配置初始词向量的第一词汇，包括：获取待分词的语料；对所述语料进行分词，得到顺序的多个分词；在所述多个分词中确定未配置初始词向量的首位分词，得到所述第一词汇。
PCT/CN2019/117725 2019-06-20 2019-11-13 词向量的配置方法、装置、存储介质、电子装置 Ceased WO2020253050A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910534810.8A CN110413990A (zh) 2019-06-20 2019-06-20 词向量的配置方法、装置、存储介质、电子装置
CN201910534810.8 2019-06-20

Publications (1)

Publication Number Publication Date
WO2020253050A1 true WO2020253050A1 (zh) 2020-12-24

Family

ID=68359467

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117725 Ceased WO2020253050A1 (zh) 2019-06-20 2019-11-13 词向量的配置方法、装置、存储介质、电子装置

Country Status (2)

Country Link
CN (1) CN110413990A (zh)
WO (1) WO2020253050A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110413990A (zh) * 2019-06-20 2019-11-05 平安科技（深圳）有限公司 词向量的配置方法、装置、存储介质、电子装置
CN113342934B (zh) * 2021-05-31 2024-04-19 北京明略软件系统有限公司 一种词向量的确定方法、装置、存储介质和电子装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150178631A1 (en) * 2013-09-04 2015-06-25 Neural Id Llc Pattern recognition system
CN108959250A (zh) * 2018-06-27 2018-12-07 众安信息技术服务有限公司 一种基于语言模型和词特征的纠错方法及其系统
CN109299269A (zh) * 2018-10-23 2019-02-01 阿里巴巴集团控股有限公司 一种文本分类方法和装置
CN110413990A (zh) * 2019-06-20 2019-11-05 平安科技（深圳）有限公司 词向量的配置方法、装置、存储介质、电子装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105608462A (zh) * 2015-12-10 2016-05-25 小米科技有限责任公司 字符相似性的判断方法及装置
CN106095865B (zh) * 2016-06-03 2019-04-12 中细软移动互联科技有限公司 一种商标文本相似性评审方法
CN109145294B (zh) * 2018-08-07 2022-05-24 北京三快在线科技有限公司 文本实体识别方法及装置、电子设备、存储介质
CN109408814B (zh) * 2018-09-30 2020-08-07 中国地质大学（武汉） 基于释义基元词的中英跨语言词汇表征学习方法及系统
CN109858039B (zh) * 2019-03-01 2023-09-05 北京奇艺世纪科技有限公司 一种文本信息识别方法及识别装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150178631A1 (en) * 2013-09-04 2015-06-25 Neural Id Llc Pattern recognition system
CN108959250A (zh) * 2018-06-27 2018-12-07 众安信息技术服务有限公司 一种基于语言模型和词特征的纠错方法及其系统
CN109299269A (zh) * 2018-10-23 2019-02-01 阿里巴巴集团控股有限公司 一种文本分类方法和装置
CN110413990A (zh) * 2019-06-20 2019-11-05 平安科技（深圳）有限公司 词向量的配置方法、装置、存储介质、电子装置

Also Published As

Publication number Publication date
CN110413990A (zh) 2019-11-05

Similar Documents

Publication Publication Date Title
US20200349175A1 (en) Address Search Method and Device
CN107436875B (zh) 文本分类方法及装置
CN105893159B (zh) 数据处理方法和装置
WO2021179701A1 (zh) 多语种语音识别方法、装置及电子设备
CN104951456B (zh) 一种用于获得答案信息的方法、装置和设备
WO2020073531A1 (zh) 文本分类模型的更新训练方法、装置及设备
CN109885828A (zh) 基于语言模型的词语纠错方法、装置、计算机设备及介质
US20220045875A1 (en) Multicast message processing method and apparatus, storage medium and processor
CN113657483A (zh) 模型训练方法、目标检测方法、装置、设备以及存储介质
CN113450805B (zh) 基于神经网络的自动语音识别方法、设备及可读存储介质
CN111310440A (zh) 文本的纠错方法、装置和系统
WO2020253506A1 (zh) 合同内容的提取方法及装置、计算机设备、存储介质
WO2020253050A1 (zh) 词向量的配置方法、装置、存储介质、电子装置
CN113127058B (zh) 数据标注方法、相关装置及计算机程序产品
CN106294340A (zh) 一种智能客服处理方法及其装置
CN109684632B (zh) 一种自然语义理解的方法、装置、计算设备
CN110162792A (zh) 电网数据管理方法及装置
CN110287318A (zh) 业务操作的检测方法及装置、存储介质、电子装置
CN110751234A (zh) Ocr识别纠错方法、装置及设备
CN113204665A (zh) 图像检索方法、装置、电子设备及计算机可读存储介质
CN117765970A (zh) 音频文件识别方法及装置
US11562011B2 (en) Incorporating data into search engines using deep learning mechanisms
CN113963360B (zh) 车牌识别方法、装置、电子设备和可读存储介质
CN103176953A (zh) 一种文本处理方法及系统
CN111291561B (zh) 文本识别方法、装置和系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19934154

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19934154

Country of ref document: EP

Kind code of ref document: A1