Information mining method, device and electronic equipment for network comments

Abstract

This disclosure provides a method, apparatus, and electronic device for information mining of online speech, relating to the field of artificial intelligence technology, specifically natural language processing, deep learning, and semantic analysis. The specific implementation scheme includes: acquiring a set of online speech samples as candidates, and mining the content semantic features and content temporal features of the online speech sample; acquiring the behavioral features of the candidate samples based on the online speech sample; predicting the probability of the candidate samples executing a target event based on the content semantic features, content temporal features, and behavioral features; and filtering out abnormal target samples from the candidate samples based on the probability of the candidate samples executing the target event. This disclosure improves the accuracy and reliability of identifying abnormal target samples by mining information from online speech, i.e., acquiring the content semantic features, content temporal features, and behavioral features of the online speech sample, and determining the probability of candidate samples executing a target event.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, specifically to the fields of natural language processing, deep learning, and semantic analysis, and particularly to a method, apparatus, and electronic storage medium for information mining of online speech. Background Technology

[0002] Faced with the vast amount of information on the internet, its relatively relaxed environment for free speech, and limited human resources, online data mining is feasible. Among related technologies, learning-to-rank (LTR) methods can be used for modeling and scoring to analyze online speech. However, LTR-based methods often rely on large amounts of data for learning, suffer from insufficient cold-start data, and have poor interpretability. Another approach is to design manually set rules based on user behavior characteristics for scoring and analysis. However, this method relies on manual rule setting, with score settings for different segments depending on business experience, requiring manual optimization and making it difficult to uncover hidden relationships between features. Therefore, improving the accuracy and reliability of online speech data mining has become an urgent problem to be solved. Summary of the Invention

[0003] This disclosure provides a method, apparatus, electronic device, storage medium, and program product for information mining of online speech.

[0004] According to the first aspect, a method for information mining of online speech is provided, comprising: acquiring a set of online speech samples of candidate objects, and mining the content semantic features and content temporal features of the online speech samples; acquiring the behavioral features of the candidate objects based on the online speech samples; predicting the probability of the candidate objects performing a target event based on the content semantic features, the content temporal features and the behavioral features; and filtering out target objects with anomalies from the candidate objects based on the probability of the candidate objects performing the target event.

[0005] According to the second aspect, an information mining device for online speech is provided, comprising: a mining module for acquiring a set of online speech samples of candidate objects and mining the set of online speech samples for content semantic features and content temporal features; an acquisition module for acquiring behavioral features of the candidate objects based on the set of online speech samples; a prediction module for predicting the probability of the candidate objects performing a target event based on the content semantic features, the content temporal features, and the behavioral features; and a filtering module for filtering out target objects with anomalies from the candidate objects based on the probability of the candidate objects performing the target event.

[0006] According to a third aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the information mining method for online speech as described in the first aspect of this disclosure.

[0007] According to a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the information mining method for online speech according to the first aspect of this disclosure.

[0008] According to a fifth aspect, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the information mining method for online speech according to the first aspect of this disclosure.

[0009] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0010] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0011] Figure 1 This is a flowchart illustrating the information mining method for online speech according to the first embodiment of this disclosure;

[0012] Figure 2 This is a flowchart illustrating the information mining method for online speech according to the second embodiment of this disclosure;

[0013] Figure 3 This is a flowchart illustrating an information mining method for online speech according to a third embodiment of this disclosure;

[0014] Figure 4 This is a flowchart illustrating the information mining method for online speech according to the fourth embodiment of this disclosure;

[0015] Figure 5 This is a flowchart illustrating the information mining method for online speech according to the fifth embodiment of this disclosure;

[0016] Figure 6 This is a flowchart illustrating the information mining method for online speech according to the sixth embodiment of this disclosure;

[0017] Figure 7This is a flowchart illustrating the information mining method for online speech according to the seventh embodiment of this disclosure;

[0018] Figure 8 This is a flowchart illustrating the information mining method for online speech disclosed in this publication;

[0019] Figure 9 This is a block diagram of an information mining apparatus for online speech used to implement embodiments of this disclosure;

[0020] Figure 10 This is a block diagram of an electronic device used to implement the information mining method for online speech in the embodiments of this disclosure. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence.

[0023] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics.

[0024] Deep learning (DL) is a new research direction in the field of machine learning (ML). It was introduced into machine learning to bring it closer to its original goal—artificial intelligence. Deep learning learns the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly helps in interpreting data such as text, images, and sound. Its ultimate goal is to enable machines to have analytical and learning capabilities like humans, and to recognize data such as text, images, and sound.

[0025] Semantic analysis is a method for analyzing semantic information based on natural language. It not only performs grammatical analysis such as lexical and syntactic analysis, but also involves the meaning contained in words, phrases, sentences, and paragraphs. Its purpose is to represent the structure of language using the semantic structure of sentences.

[0026] The following describes an embodiment of the information mining method for online speech based on the accompanying drawings.

[0027] Figure 1 This is a flowchart illustrating the information mining method for online speech according to the first embodiment of this disclosure.

[0028] like Figure 1 As shown, the information mining method for online speech in this embodiment of the present disclosure may specifically include the following steps:

[0029] S101: Obtain the online opinion set of the candidate objects, and mine the content semantic features and content temporal features of the online opinion set.

[0030] Specifically, the execution entity of the information mining method for online speech in this disclosure embodiment can be the information mining device for online speech provided in this disclosure embodiment. This information mining device for online speech can be a hardware device with data processing capabilities and / or the necessary software to drive the hardware device. Optionally, the execution entity may include a workstation, server, computer, user terminal, and other devices. The user terminal includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, and vehicle terminals.

[0031] It should be noted that this disclosure does not limit the specific method of obtaining the online opinion collection of candidate candidates, and the selection can be made according to the actual situation.

[0032] Optionally, a collection of online comments can be obtained by collecting online comments generated by users while using different applications.

[0033] Optionally, the time for collecting online opinions can be set.

[0034] For example, we can collect online comments generated by users during their use of different applications over the past week to obtain a collection of online comments, or we can collect online comments generated by users during their use of different applications over the past month to obtain a collection of online comments.

[0035] In this embodiment of the disclosure, after obtaining the set of online opinions, the online opinions in the set can be preprocessed to obtain preprocessed online opinions, and the content semantic features and content temporal features of the preprocessed online opinions can be mined.

[0036] For example, since there are high-frequency words in online speech texts that have no real meaning, these words are useless for mining the semantic and temporal features of online speech datasets. Useless words and colloquial words, such as "ba," "ah," and "uh," can be removed from online speech texts. Removing these words can reduce the noise in the online speech text dataset.

[0037] It should be noted that if the semantic analysis is based solely on the number of words in the sensitive vocabulary list that the text contains, without considering the meaning and context of the entire sentence, it can lead to incorrect analysis results. Online comments posted by users are often highly time-sensitive, and the more recent the comments, the more they reflect the user's current psychological state and behavioral tendencies. After obtaining a set of online comments, the semantic features and temporal features of the content can be mined.

[0038] S102, Based on the online speech set, obtain the behavioral characteristics of the candidate objects.

[0039] In this embodiment of the disclosure, after obtaining the set of online opinions, the behavioral characteristics of candidate objects can be obtained by calculation based on the set of online opinions.

[0040] Optionally, features can be constructed from the search behavior information of candidate objects based on a set of online comments to obtain the behavioral features of candidate objects.

[0041] Optionally, the behavioral characteristics of the candidates can be search frequency, number of active days, time since the last active time, duration, number of times specific keyword behavior tags are hit, etc.

[0042] S103, based on the semantic features, temporal features, and behavioral features of the content, predict the probability of the candidate object executing the target event.

[0043] It should be noted that this disclosure does not limit the specific method for predicting the probability of a candidate object executing a target event based on content semantic features, content temporal features, and behavioral features; the appropriate method can be selected based on the actual situation.

[0044] Optionally, scoring can be performed based on a Gradient Boosting Decision Tree (GBDT) model. The GBDT model outputs a score, and the probability of a candidate object executing the target event is generated based on the score.

[0045] It should be noted that this disclosure does not impose any restrictions on the settings of the target event, and the settings can be made according to the actual situation.

[0046] S104. Based on the probability of the candidate object executing the target event, filter out the target object that is abnormal from the candidate objects.

[0047] Optionally, a probability threshold can be set for candidate objects to execute the target event, and target objects with anomalies can be filtered out from the candidate objects based on the probability of candidate objects executing the target event and the probability threshold.

[0048] For example, if the probability of a candidate object executing the target event is greater than or equal to a probability threshold, then the candidate object can be determined to be the target object; if the probability of a candidate object executing the target event is less than the probability threshold, then the candidate object can be determined to be an object without any anomalies.

[0049] Optionally, a probability threshold for candidate objects to execute the target event can be set, and the probabilities of candidate objects executing the target event can be sorted. Based on the probability of candidate objects executing the target event, the probability threshold, and the sorting results, target objects with anomalies can be filtered out from the candidate objects.

[0050] Furthermore, after filtering out the target objects that are abnormal, the target objects can be pushed out one by one.

[0051] Optionally, the probability of executing the target event on the candidate objects can be sorted from largest to smallest, and the target objects with the highest ranking can be pushed in sequence according to the ranking result.

[0052] In summary, the information mining method for online speech in this disclosure obtains a set of online speech samples from candidate objects, mines the content semantic features and content temporal features of the online speech samples, obtains the behavioral features of the candidate objects based on the online speech samples, predicts the probability of the candidate objects performing a target event based on the content semantic features, content temporal features, and behavioral features, and filters out target objects with abnormalities from the candidate objects based on the probability of the candidate objects performing the target event. This disclosure improves the accuracy and reliability of identifying target objects with abnormalities by mining information from online speech, namely obtaining the content semantic features, content temporal features, and behavioral features of the online speech samples, and determining whether there are abnormal target objects based on the probability of the candidate objects performing the target event.

[0053] Figure 2 This is a flowchart illustrating an information mining method for online speech according to a second embodiment of this disclosure.

[0054] like Figure 2 As shown, in Figure 1 Based on the illustrated embodiments, the information mining method for online speech in this disclosure may specifically include the following steps:

[0055] S201, Obtain the online opinion set of the candidate object, and mine the content semantic features and content temporal features of the online opinion set.

[0056] S202, Based on online speech sets, obtain the behavioral characteristics of candidate objects.

[0057] Specifically, steps S201-S202 in this embodiment are the same as steps S101-S102 in the above embodiment, and will not be repeated here.

[0058] The step S103 in the above embodiment, "predicting the probability of a candidate object executing a target event based on content semantic features, content temporal features, and behavioral features," may specifically include the following steps S203 to S205.

[0059] S203, based on the semantic features of the mined content, determine the first and second quantitative features of the candidate objects' tendency to perform the target event.

[0060] As one possible implementation, such as Figure 3 As shown, based on the above embodiments, the specific process of determining the first and second quantitative features of the candidate object's tendency to execute the target event according to the mined content semantic features in step S202 includes the following steps:

[0061] S301, Based on the semantic features of the mined content, determine the category confidence of each tendency category under the candidate object's tendency to execute the target event, wherein the tendency categories are divided according to the degree to which the candidate object tends to execute the target event.

[0062] For example, when the target event is a candidate purchasing product A, a category confidence score is constructed for each tendency category for this business scenario. Taking the target event "a candidate purchasing product A" as an example for explanation, the degree of tendency to perform the target event is "high", which clearly shows the intention to purchase A. There are often words related to "purchase", reflecting a strong idea of ​​purchasing or that it is very likely that the purchase has already been made. The degree of tendency to perform the target event is "medium", which means that it is not completely certain whether there is a tendency to purchase product A. The degree of tendency to perform the target event is "low", which means that the intention to purchase product A can be completely ruled out.

[0063] Furthermore, the degree of tendency to perform the target event can be labeled, with "high" labeled as 2, "medium" as 1, and "low" as 0. The training set and validation set can be divided, with 70% of the labeled data used as the training set and 30% as the validation set.

[0064] Alternatively, the class confidence score for each tendency category of the candidate object to perform the target event can be determined according to the following formula, i.e., the class confidence scores for high (2)-medium (1)-low (0):

[0065] vector[confidence] 0,, =softmax(W T x+b)

[0066] S302, determine the first quantitative feature and the second quantitative feature based on the category confidence of online speech under different tendency categories.

[0067] As one possible implementation, such as Figure 4 As shown, based on the above embodiments, the specific process of determining the first quantitative feature in step S302 according to the category confidence of online speech under different tendency categories includes the following steps:

[0068] S401, Obtain the category confidence of each online comment in the online comment set under the target tendency category.

[0069] It should be noted that the confidence level of each online opinion in the online opinion set under the target tendency category can be obtained, that is, the confidence level of the category under the three tendency categories of high (2)-medium (1)-low (0).

[0070] S402, from the category confidence scores under the target tendency category, count the first number of category confidence scores that are greater than the set confidence score threshold.

[0071] It should be noted that the presence of "high" or "low" confidence scores can significantly affect the assessment of abnormal targets. At the same time, there are many "medium" scores, which can be ambiguous and difficult to distinguish abnormal targets. Therefore, we can count the number of categories with "high" and "low" confidence scores that are greater than the set confidence threshold from the category confidence scores under the target's tendency category.

[0072] S403, determine the first total number of online comments.

[0073] In this embodiment of the disclosure, after obtaining the set of online opinions, the first total number of online opinions in the set of online opinions can be determined.

[0074] S404, the proportion of high confidence is determined based on the first quantity and the first total quantity of online comments, and is used as the first quantitative feature.

[0075] Optionally, after obtaining the first number and the first total number of online comments, the proportion of high confidence can be determined according to the following formula:

[0076]

[0077] As one possible implementation, such as Figure 5 As shown, based on the above embodiments, the specific process of determining the second quantitative feature in step S302 according to the category confidence of online speech under different tendency categories includes the following steps:

[0078] S501, determine the difference between the category confidence scores of pairwise tendency categories of online speech.

[0079] It should be noted that after obtaining the category confidence scores under different tendency categories, the difference between the category confidence scores of pairwise tendency categories of online speech can be calculated.

[0080] S502, obtain the first total number of online comments in the online comment collection.

[0081] S503, the confidence mean difference is determined based on the difference in the target network statements and the first total number, and is used as the second quantitative feature.

[0082] Optionally, after obtaining the difference and the first total number of target online comments, the mean confidence level can be determined according to the following formula:

[0083]

[0084] S204. Based on the temporal characteristics of the mined content, determine the third quantitative characteristic of the candidate object's tendency to execute the target event.

[0085] As one possible implementation, such as Figure 6 As shown, based on the above embodiments, the specific process of determining the third quantitative feature of the candidate object's tendency to execute the target event according to the temporal characteristics of the mined content in step S204 includes the following steps:

[0086] S601, sort the online speech set according to time sequence and truncate the tail to obtain the online speech sequence.

[0087] Optionally, the set of online comments can be sorted chronologically, and the last one-sixth can be truncated to obtain the sequence of online comments.

[0088] S602, obtain the category confidence of each online statement in the online statement sequence under each category.

[0089] For example, for a network sequence of [2, 2, 0, 0, 0, 0], the class confidence under each category is [high, high, low, low, low].

[0090] S603, based on the category confidence of each category of online speech, determines the proportion of online speech in the low-bias category in the online speech sequence, as the third quantitative feature of the candidate object.

[0091] As one possible implementation, such as Figure 7 As shown, based on the above embodiments, the specific process of determining the proportion of low-biased opinions in the online opinion sequence according to the category confidence level in step S603 includes the following steps:

[0092] S701. Select the category with the highest category confidence among all categories as the identification tendency type of online speech.

[0093] Optionally, online comments can be sorted chronologically to obtain a sequence over a period of time. The category with the highest confidence level among all categories is then selected as the type of online commentary to identify its tendency. That is, the category with the highest confidence level can be taken as the sequence value.

[0094] val i =argmax(vector[])

[0095] S702, the second number of online comments in a statistical online commentary sequence that are identified as having a low-tendency type.

[0096] In this embodiment of the disclosure, after obtaining the online speech sequence, a second number of online speech in the online speech sequence that is identified as having a low tendency type can be counted.

[0097] S703, obtain the second total number of online comments in the online comment sequence.

[0098] In this embodiment of the disclosure, after obtaining the online speech sequence, a second total number of online speech in the online speech sequence can be counted to obtain the second total number of online speech in the online speech sequence.

[0099] S704. Determine the proportion of low-biased online comments in the online comment sequence based on the second number and the second total number of low-biased online comments.

[0100] Optionally, after obtaining the second quantity and the second total quantity, the proportion of low-biased comments in the online comment sequence can be determined according to the following formula:

[0101]

[0102] For example: Given a network sequence [2, 2, 0, 0, 0, 0], which is already sorted chronologically, and setting the recent time point parameter to 1 / 6, the last 1 / 6 of the sequence is taken as the most recent network sequence. If the percentage of low-biased comments is calculated to be 1, then the output value is 1.

[0103] S205, based on behavioral characteristics, first quantitative characteristics, second quantitative characteristics and third quantitative characteristics, predict the probability of the candidate object executing the target event, and obtain the probability of the candidate object executing the target event.

[0104] Optionally, after obtaining the behavioral features, the first quantization feature, the second quantization feature, and the third quantization feature, a feature vector can be constructed. The constructed feature vector is used as the input of the gradient boosting decision tree model to predict the probability of the candidate object executing the target event based on the gradient boosting decision tree model, thereby obtaining the probability of the candidate object executing the target event.

[0105] S206, Based on the probability that the candidate object will execute the target event, filter out the target objects that are abnormal from the candidate objects.

[0106] Specifically, step S206 in this embodiment is the same as step S104 in the above embodiment, and will not be repeated here.

[0107] In summary, the information mining method for online speech in this disclosure obtains a set of online speech samples from candidate objects, mines the content semantic features and content temporal features of the online speech samples, obtains the behavioral features of the candidate objects based on the online speech samples, and determines a first quantitative feature and a second quantitative feature of the candidate objects' tendency to perform target events based on the mined content semantic features. Based on the mined content temporal features, a third quantitative feature of the candidate objects' tendency to perform target events is determined. The probability of the candidate objects performing target events is predicted based on the behavioral features, the first quantitative feature, the second quantitative feature, and the third quantitative feature, thus obtaining the probability of the candidate objects performing target events. Based on the probability of the candidate objects performing target events, target objects with abnormalities are screened out from the candidate objects. Therefore, this disclosure improves the accuracy and reliability of identifying target objects with abnormalities by mining information from online speech, i.e., obtaining the behavioral features, the first quantitative feature, the second quantitative feature, and the third quantitative feature of the online speech samples, and determining the probability of candidate objects performing target events.

[0108] The following section explains the methods for information mining of online speech.

[0109] For example, such as Figure 8As shown, firstly, the text corresponding to the online comments is input and preprocessed, such as removing word segmentation and stop words. Then, the semantic features of the content are mined, namely, the first quantitative feature is the proportion of high confidence, the second quantitative feature is the average difference of confidence, and based on the mined temporal features of the content, the third quantitative feature is determined to indicate the candidate object's tendency to perform the target event. The behavioral features of the candidate objects are also obtained. By constructing a feature vector based on the behavioral features, the first quantitative feature, the second quantitative feature, and the third quantitative feature, the probability of the candidate object performing the target event is predicted, thus obtaining the probability of the candidate object performing the target event. Furthermore, based on the probability of the candidate object performing the target event, the target objects with abnormalities can be screened out from the candidate objects.

[0110] In summary, the information mining method for online speech in this disclosure obtains a set of online speech samples from candidate objects, mines the content semantic features and content temporal features of the online speech samples, obtains the behavioral features of candidate objects based on the online speech samples, and determines a first quantitative feature and a second quantitative feature of candidate objects' tendency to perform target events based on the mined content semantic features. Based on the mined content temporal features, a third quantitative feature of candidate objects' tendency to perform target events is determined. The probability of candidate objects performing target events is predicted based on the behavioral features, the first quantitative feature, the second quantitative feature, and the third quantitative feature, thus obtaining the probability of candidate objects performing target events. Based on the probability of candidate objects performing target events, abnormal target objects are screened out from the candidate objects. Therefore, this disclosure improves the accuracy and reliability of identifying abnormal target objects by mining information from online speech samples—that is, obtaining the behavioral features, the first quantitative feature, the second quantitative feature, and the third quantitative feature of the online speech samples—and determining whether abnormal target objects exist based on the probability of candidate objects performing target events. This increases the positive feedback rate of users and the hit rate of target object judgment.

[0111] It should be noted that the acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0112] Figure 9 This is a schematic diagram of a device for mining online speech according to an embodiment of the present disclosure.

[0113] like Figure 9 As shown, the information mining device 900 for online speech includes: a mining module 910, an acquisition module 920, a prediction module 930, and a filtering module 940. Wherein:

[0114] The mining module 910 is used to obtain a set of online opinions of candidate objects and to mine the set of online opinions by means of semantic features and temporal features.

[0115] The acquisition module 920 is used to acquire the behavioral characteristics of the candidate objects based on the network speech set;

[0116] The prediction module 930 is used to predict the probability that the candidate object will perform the target event based on the content semantic features, the content temporal features, and the behavioral features.

[0117] The filtering module 940 is used to filter out target objects that are abnormal from the candidate objects based on the probability that the candidate objects will perform the target event.

[0118] The prediction module 930 is also used for:

[0119] Based on the mined semantic features of the content, a first quantitative feature and a second quantitative feature are determined to indicate that the candidate object is inclined to perform the target event;

[0120] Based on the temporal characteristics of the mined content, a third quantitative characteristic is determined for the candidate object to tend to execute the target event;

[0121] Based on the behavioral characteristics, the first quantitative characteristics, the second quantitative characteristics, and the third quantitative characteristics, the probability of the candidate object executing the target event is predicted, thereby obtaining the probability of the candidate object executing the target event.

[0122] The prediction module 930 is also used for:

[0123] Based on the mined semantic features of the content, the category confidence of each tendency category of the candidate object to perform the target event is determined, wherein the tendency categories are divided according to the degree to which the candidate object tends to perform the target event;

[0124] The first quantitative feature and the second quantitative feature are determined based on the category confidence of the online comments under different tendency categories.

[0125] The prediction module 930 is also used for:

[0126] Obtain the category confidence score of each online comment in the online comment set under the target tendency category;

[0127] From the category confidence scores under the target tendency category, count the first number of category confidence scores that are greater than the set confidence threshold;

[0128] Determine the first total number of online comments in the aforementioned online commentary set;

[0129] The high confidence percentage is determined based on the first quantity and the first total quantity of online comments, and is used as the first quantitative feature.

[0130] The prediction module 930 is also used for:

[0131] Determine the difference between the category confidence scores of each pair of tendency categories of the online speech;

[0132] Obtain the first total number of online comments in the aforementioned online commentary set;

[0133] The confidence level difference is determined based on the difference between the target network statements and the first total number, and is used as the second quantitative feature.

[0134] The prediction module 930 is also used for:

[0135] The online opinion set is sorted and its tail is truncated according to time sequence to obtain the online opinion sequence.

[0136] Obtain the category confidence score of each online comment in the online comment sequence under each category;

[0137] Based on the category confidence level under each category of the online speech, the proportion of online speech in the low-bias category in the online speech sequence is determined as the third quantitative feature of the candidate object.

[0138] The prediction module 930 is also used for:

[0139] The category with the highest confidence level among all categories is selected as the identification tendency type of the online speech;

[0140] The second number of online comments in the aforementioned online comment sequence that are classified as low-tendency type is counted.

[0141] Obtain the second total number of online comments in the online comment sequence;

[0142] The proportion of low-biased online comments in the online comment sequence is determined based on the second quantity of low-biased online comments and the second total quantity.

[0143] Device 900 is also used for:

[0144] The online comments in the aforementioned online comment set are preprocessed to obtain preprocessed online comments, and feature mining is performed on the preprocessed online comments.

[0145] It should be noted that the above explanation of the embodiments of the information mining method for online speech also applies to the information mining apparatus for online speech in the embodiments of this disclosure, and the specific process will not be repeated here.

[0146] In summary, the information mining apparatus for online speech according to this disclosure acquires a set of online speech samples from candidate objects, mines the content semantic features and content temporal features of the online speech samples, obtains the behavioral features of the candidate objects based on the online speech samples, predicts the probability of the candidate objects performing a target event based on the content semantic features, content temporal features, and behavioral features, and filters out target objects with abnormalities from the candidate objects based on the probability of the candidate objects performing the target event. This disclosure improves the accuracy and reliability of identifying target objects with abnormalities by mining information from online speech, namely, acquiring the content semantic features, content temporal features, and behavioral features of the online speech samples, and determining whether there are abnormal target objects based on the probability of the candidate objects performing the target event.

[0147] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0148] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0149] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded from storage unit 1007 into random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0150] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0151] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as information mining methods for online speech. For example, in some embodiments, the information mining method for online speech can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the information mining method for online speech described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured in any other suitable manner (e.g., by means of firmware) to perform information mining methods for online speech.

[0152] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0156] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0157] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0158] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the information mining method for online speech as described above.

[0159] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0160] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A network opinion-oriented information mining method, wherein, The method comprises the following steps: obtaining a network comment set of a candidate object, and performing content semantic feature and content timing feature mining on the network comment set; obtaining behavior features of the candidate object based on the network comment set; determining a category confidence of each tendency category in which the candidate object tends to perform a target event according to the mined content semantic features, wherein the tendency categories are divided according to the degree of the candidate object's tendency to perform the target event; determining a first quantitative feature and a second quantitative feature according to the category confidence of the network comment in different tendency categories; sorting and tail cutting the network comment set according to the time sequence to obtain a network comment sequence; obtaining the category confidence of each network comment in each category in the network comment sequence; determining the proportion of network comments of a low tendency category in the network comment sequence as a third quantitative feature of the candidate object according to the category confidence of each category of the network comment; predicting the probability of the candidate object performing the target event according to the behavior features, the first quantitative feature, the second quantitative feature and the third quantitative feature; screening out a target object with an anomaly from the candidate object according to the probability of the candidate object performing the target event.

2. The method of claim 1, wherein, According to the category confidence of the network comment in different tendency categories, the first quantitative feature is determined, comprising: obtaining the category confidence of each network comment in a target tendency category in the network comment set; from the category confidence in the target tendency category, the first number of category confidences greater than a set confidence threshold is counted; determining the first total number of network comments in the network comment set; determining a high confidence proportion according to the first number and the first total number of network comments as the first quantitative feature.

3. The method of claim 1, wherein, According to the category confidence of the network comment in different tendency categories, the second quantitative feature is determined, comprising: determining the difference between the category confidences of each pair of tendency categories of the network comment; obtaining the first total number of network comments in the network comment set; determining a confidence mean difference according to the difference of the network comment and the first total number as the second quantitative feature.

4. The method of claim 1, wherein, According to the category confidence, the proportion of low tendency comments in the network comment sequence is determined, comprising: selecting the category with the highest category confidence among all categories as the identified tendency type of the network comment; counting the second number of network comments of the identified tendency type as a low tendency type in the network comment sequence; obtaining the second total number of network comments in the network comment sequence; determining the proportion of low tendency comments in the network comment sequence according to the second number of network comments of the low tendency type and the second total number.

5. The method of claim 1, wherein, The method further comprises: preprocessing the network comments in the network comment set to obtain preprocessed network comments, and performing feature mining on the preprocessed network comments.

6. A network opinion-oriented information mining apparatus, wherein, The method comprises the following steps: a mining module for obtaining a network comment set of a candidate object, and performing content semantic feature and content timing feature mining on the network comment set; The acquisition module is configured to acquire a behavior feature of the candidate object based on the set of network comments. The prediction module is configured to determine a category confidence of each tendency category in which the candidate object tends to perform a target event according to the mined content semantic features, wherein the tendency categories are divided according to degrees in which the candidate object tends to perform the target event; determine a first quantitative feature and a second quantitative feature according to the category confidence of the network comments in different tendency categories; The network comment set is sorted and tail-truncated according to time sequence to obtain a network comment sequence; the category confidence of each network comment in each category in the network comment sequence is acquired; and a proportion of network comments in a low tendency category in the network comment sequence is determined as a third quantitative feature of the candidate object according to the category confidence of each network comment in each category. The prediction module is further configured to: Acquire the category confidence of each network comment in the target tendency category in the set of network comments; 7. The apparatus of claim 6, wherein, Count a first number of category confidences greater than a set confidence threshold from the category confidences in the target tendency category; Determine a first total number of network comments in the set of network comments; Determine a high confidence proportion according to the first number and the first total number of network comments as the first quantitative feature. The prediction module is further configured to: Determine a difference between the category confidences of each pair of tendency categories of the network comments; 8. The apparatus of claim 6, wherein, Acquire a first total number of network comments in the set of network comments; Determine a confidence mean difference according to the difference of the network comments and the first total number as the second quantitative feature. The prediction module is further configured to: Select a category with the highest category confidence among all categories as an identified tendency type of the network comments; 9. The apparatus of claim 6, wherein, Count a second number of network comments of the identified tendency type in the network comment sequence that are in a low tendency category; Acquire a second total number of network comments in the network comment sequence; Determine a proportion of network comments in a low tendency category in the network comment sequence according to the second number of network comments of the low tendency category and the second total number. The device is further configured to: Preprocess the network comments in the set of network comments to obtain preprocessed network comments, and mine features from the preprocessed network comments.

10. The apparatus of claim 9, wherein, The device includes a processor and a memory; The processor reads executable program codes stored in the memory to run programs corresponding to the executable program codes to implement the method of any one of claims 1-5.

11. An electronic device, comprising: The program is executed by the processor to implement the method of any one of claims 1-5. ​ 12. A computer readable storage medium having stored thereon a computer program, characterized in that, ​ 13. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-5.

Images