CN111708866A - Session segmentation method and device, electronic equipment and storage medium - Google Patents
Session segmentation method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111708866A CN111708866A CN202010853510.9A CN202010853510A CN111708866A CN 111708866 A CN111708866 A CN 111708866A CN 202010853510 A CN202010853510 A CN 202010853510A CN 111708866 A CN111708866 A CN 111708866A
- Authority
- CN
- China
- Prior art keywords
- segmentation
- session
- message
- adjacent
- messages
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The application provides a session segmentation method, a session segmentation device, electronic equipment and a storage medium, wherein the session segmentation method comprises the following steps: acquiring adjacent messages to be processed in a target session; under the condition that the adjacent messages accord with preset segmentation conditions, determining to segment the adjacent messages; under the condition that the adjacent messages accord with the preset continuous conditions, determining not to split the adjacent messages; and under the condition that the adjacent messages do not accord with the segmentation condition and the consistency condition, determining whether to segment the adjacent messages according to the session segmentation model. The method and the device can improve the accuracy of session segmentation.
Description
Technical Field
The present application relates to the field of data processing, and in particular, to a session segmentation method and apparatus, an electronic device, and a storage medium.
Background
With the development of intelligent terminals and mobile communication technologies, more and more users use electronic devices to send and receive messages, and mutual conversation is achieved. The electronic equipment is used for conversation, the functions of the electronic equipment in the fields of life information intercommunication sharing, remote cooperative office, team daily management and the like are gradually highlighted, the communication efficiency of people in life and work can be improved, and the electronic equipment is convenient for people to store historical conversation information.
Based on this, in the related art, the information required by people can be acquired by using the historical session information in the electronic equipment. In order to obtain specific information, the sessions need to be segmented so as to analyze the sessions with different topics and extract information. However, in the related art, the way of segmenting the session has the problem of inaccurate segmentation.
Disclosure of Invention
The embodiment of the application provides a session segmentation method, a session segmentation device, electronic equipment and a storage medium, which are used for solving the problems in the related art, and the technical scheme is as follows:
in a first aspect, an embodiment of the present application provides a session segmentation method, including:
acquiring adjacent messages to be processed in a target session;
under the condition that the adjacent messages accord with preset segmentation conditions, determining to segment the adjacent messages;
under the condition that the adjacent messages accord with the preset continuous conditions, determining not to split the adjacent messages;
and under the condition that the adjacent messages do not accord with the segmentation condition and the consistency condition, determining whether to segment the adjacent messages according to the session segmentation model.
In one embodiment, the method further comprises:
and training to obtain a session segmentation model by using the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session.
In one embodiment, the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session comprises:
acquiring an initial model obtained based on session scene corpus training;
and training the initial model by using the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session to obtain a session segmentation model.
In one embodiment, the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session comprises:
connecting adjacent messages meeting the segmentation conditions in the target session with at least one message in front of the adjacent messages to obtain first training data marked as segmentation;
connecting the adjacent message meeting the consistency condition in the target session with at least one message in front of the adjacent message to obtain second training data marked as non-segmentation;
the session segmentation model is trained using the first training data and the second training data.
In one embodiment, the cutting conditions include at least one of the following conditions:
the speakers of adjacent messages belong to the same role, and adjacent messages are sent out on different dates;
the sending time interval of the adjacent messages is greater than the preset time length, and the message with later sending time in the adjacent messages comprises a preset starting language;
sending out adjacent messages on different dates, wherein the messages with earlier sending time in the adjacent messages comprise preset end words;
the adjacent messages are sent out on different dates, and the type of the message with the earlier sending time in the adjacent messages is a webpage link;
the type of the message with later sending time in the adjacent messages is a notice;
the type of the message with the later sending time in the adjacent messages is a webpage link or a number, and the sending time interval of the adjacent messages is more than the preset time length.
In one embodiment, the consistent conditions include at least one of the following:
the message with later sending time in the adjacent messages is an empty message;
the message with later sending time consists of emoticons and long texts; the long text is a text with more words than a preset threshold value;
messages sent later in time consist of numbers;
messages that are sent later in time are composed of person names and non-text messages.
In one embodiment, the method further comprises:
obtaining a predicted segmentation position in the target session according to the segmentation result of the adjacent messages;
calculating to obtain a first evaluation index according to the predicted segmentation position and a pre-marked actual segmentation position; wherein the first evaluation index is used for adjusting at least one of a segmentation condition, a coherence condition and a session segmentation model.
In one embodiment, the method further comprises:
segmenting the target session into a plurality of predicted session segments according to the predicted segmentation positions;
calculating to obtain a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; and the second evaluation index is used for evaluating the segmentation quality of the target session.
In a second aspect, an embodiment of the present application provides a session splitting apparatus, including
The acquisition module is used for acquiring adjacent messages to be processed in the target session;
the segmentation module is used for determining to segment the adjacent messages under the condition that the adjacent messages meet the preset segmentation conditions;
the consistency module is used for determining not to segment the adjacent messages under the condition that the adjacent messages accord with the preset consistency condition;
and the determining module is used for determining whether to segment the adjacent messages according to the session segmentation model under the condition that the adjacent messages do not accord with the segmentation condition and the consistency condition.
In one embodiment, the apparatus further comprises:
and the training module is used for training to obtain the session segmentation model by utilizing the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session.
In one embodiment, the training module comprises:
the initialization unit is used for acquiring an initial model obtained based on session scene corpus training;
and the first training unit is used for training the initial model by utilizing the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target conversation to obtain a conversation segmentation model.
In one embodiment, the training module comprises:
the first connecting unit is used for connecting the adjacent message meeting the segmentation condition in the target session with at least one message before the adjacent message to obtain first training data marked as segmentation;
the second connecting unit is used for connecting the adjacent message meeting the consistency condition in the target session with at least one message before the adjacent message to obtain second training data marked as non-segmentation;
the second training unit trains the session segmentation model using the first training data and the second training data.
In one embodiment, the cutting conditions include at least one of the following conditions:
the speakers of adjacent messages belong to the same role, and adjacent messages are sent out on different dates;
the sending time interval of the adjacent messages is greater than the preset time length, and the message with later sending time in the adjacent messages comprises a preset starting language;
sending out adjacent messages on different dates, wherein the messages with earlier sending time in the adjacent messages comprise preset end words;
the adjacent messages are sent out on different dates, and the type of the message with the earlier sending time in the adjacent messages is a webpage link;
the type of the message with later sending time in the adjacent messages is a notice;
the type of the message with the later sending time in the adjacent messages is a webpage link or a number, and the sending time interval of the adjacent messages is more than the preset time length.
In one embodiment, the consistent conditions include at least one of the following:
the message with later sending time in the adjacent messages is an empty message;
the message with later sending time consists of emoticons and long texts; the long text is a text with more words than a preset threshold value;
messages sent later in time consist of numbers;
messages that are sent later in time are composed of person names and non-text messages.
In one embodiment, the apparatus further comprises:
the segmentation position module is used for obtaining a prediction segmentation position in the target session according to the segmentation result of the adjacent message;
the first evaluation module is used for calculating to obtain a first evaluation index according to the predicted segmentation position and the pre-marked actual segmentation position; wherein the first evaluation index is used for adjusting at least one of a segmentation condition, a coherence condition and a session segmentation model.
In one embodiment, the apparatus further comprises:
the session segment module is used for segmenting the target session into a plurality of predicted session segments according to the predicted segmentation positions;
the second evaluation module is used for calculating to obtain a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; and the second evaluation index is used for evaluating the segmentation quality of the target session.
In a third aspect, an embodiment of the present application provides an electronic device, including: a memory and a processor. Wherein the memory and the processor are in communication with each other via an internal connection path, the memory is configured to store instructions, the processor is configured to execute the instructions stored by the memory, and the processor is configured to perform the method of any of the above aspects when the processor executes the instructions stored by the memory.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, which stores a computer program, and when the computer program runs on a computer, the method in any one of the above-mentioned aspects is executed.
The advantages or beneficial effects in the above technical solution at least include:
according to the technical scheme, whether the adjacent messages in the target session need to be segmented or not can be judged based on the preset segmentation condition and the preset consistency condition. And under the condition that whether the adjacent messages need to be segmented cannot be determined based on the preset segmentation condition and the preset coherence condition, determining whether to segment the adjacent messages according to a session segmentation model. Therefore, the accuracy of segmenting the target session can be improved by setting the segmentation condition and the consistency condition, the misjudgment of adjacent messages which do not accord with the segmentation condition or the consistency condition is avoided by using the session segmentation model, and the accuracy of segmenting the target session is further improved.
The foregoing summary is provided for the purpose of description only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present application will be readily apparent by reference to the drawings and following detailed description.
Drawings
In the drawings, like reference numerals refer to the same or similar parts or elements throughout the several views unless otherwise specified. The figures are not necessarily to scale. It is appreciated that these drawings depict only some embodiments in accordance with the disclosure and are therefore not to be considered limiting of its scope.
Fig. 1 is a schematic diagram of a session segmentation method according to an embodiment of the present application;
fig. 2 is a schematic diagram of a session segmentation method according to another embodiment of the present application;
fig. 3 is a schematic diagram of a session segmentation method according to another embodiment of the present application;
FIG. 4 is a schematic diagram of an exemplary application of the present application;
FIG. 5 is a schematic diagram of another example of an application of the present application;
fig. 6 is a block diagram illustrating a structure of a session segmentation apparatus according to an embodiment of the present application;
fig. 7 is a block diagram illustrating a structure of a session segmentation apparatus according to another embodiment of the present application;
fig. 8 is a block diagram of an electronic device for implementing the session segmentation method of the present application.
Detailed Description
In the following, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present application. Accordingly, the drawings and description are to be regarded as illustrative in nature, and not as restrictive.
Fig. 1 shows a flowchart of a session segmentation method according to an embodiment of the present application. As shown in fig. 1, the method may include:
step S11, acquiring the adjacent message to be processed in the target session;
step S12, determining to segment the adjacent messages under the condition that the adjacent messages meet the preset segmentation conditions;
step S13, under the condition that the adjacent message meets the preset continuous condition, determining not to split the adjacent message;
and step S14, under the condition that the adjacent messages do not accord with the segmentation condition and the consistency condition, determining whether to segment the adjacent messages according to the conversation segmentation model.
In the embodiment of the present application, the target session may refer to session information to be split. Illustratively, session information, such as personal chat information or group session information, obtained by a user sending and receiving information to and from other users using an instant messaging program on the electronic device may be included.
Illustratively, adjacent messages can be sequentially extracted from all messages of the target session as the adjacent messages to be processed. For example, first extract the 1 st message and the 2 nd message in the target session, then extract the 2 nd message and the 3 rd message, then extract the 3 rd message and the 4 th message, and so on, extract all the messages of the target session. One set of messages extracted at a time may be treated as a neighbor message to be processed.
Or, the messages in the target session may be screened to obtain the neighbor messages to be processed. For example, messages of the target session are screened for neighboring messages sent on alternate days. If the former message in a group of adjacent messages in the target conversation is sent out on the 1 st day of the conversation and the latter message is sent out on the 2 nd day of the conversation, the group of adjacent messages are adjacent messages to be processed; if another group of adjacent messages in the target conversation are all sent on the 2 nd day of the conversation, the group of adjacent messages are not processed, for example, it can be directly determined that the group of adjacent messages do not need to be split.
In the embodiment of the present application, the adjacent messages are segmented, which means that a conversation is segmented between two adjacent messages, for example, the conversation is segmented between adjacent messages "good" and "good" so that "good" is the last message of the previous conversation segment and "good" is the first message of the next conversation segment.
The splitting condition corresponds to that adjacent messages can be split, and may include that the sending time interval of the adjacent messages exceeds a preset threshold, the previous message in the adjacent messages includes an end word such as "next week", "like this bar" or the like, or the next message includes a start word such as "at do", "you do", or the like, so that the adjacent messages meeting the splitting condition do not belong to the same session segment. The consistency condition corresponds to no slicing of adjacent messages, and may include that a previous message in adjacent messages includes a start word or that a subsequent message is a null message, etc. By setting the segmentation condition and the consistency condition, the accuracy of session segmentation can be controlled. For example, making the conditions more stringent may improve the accuracy of session slicing.
For example, the time information, the content information and the role information of two messages can be extracted for each group of adjacent messages. The time information includes, for example, sending time, the content information includes, for example, a content type, whether a preset keyword or phrase exists in the content, and the like, and the role information is, for example, a teacher, a student, a parent, and the like in an education group session, or a speaker role in a session of another scene. And then, determining whether the adjacent messages meet the preset segmentation condition and the preset continuity condition or not according to the time information, the content information and/or the role information of the adjacent messages.
Alternatively, the slicing conditions may include one or more of various conditions in the following examples:
example one: the segmentation condition determined according to the time and the role information comprises the following steps:
the speakers of adjacent messages belong to the same role and adjacent messages are sent out on different dates.
Example two: the segmentation condition determined according to the time and the content information comprises the following steps:
the sending time interval of the adjacent messages is larger than a preset time length, such as 2 minutes, 4 minutes and the like, and the message sent later in the adjacent messages comprises a preset starting word, such as 'Domo', 'hello' and the like;
the adjacent messages are sent out on different dates, and the message sent out later in time in the adjacent messages comprises preset starting words such as 'how recent' and the like;
adjacent messages are sent out on different dates, and messages sent out earlier in time in the adjacent messages comprise preset end words such as 'good', 'next week' and the like; alternatively, it may be defined that the message with the earlier issue time includes only short text, for example, text with a word count less than 3, in addition to the end word;
adjacent messages are sent on different dates and the type of message sent earlier in the adjacent message is a web page link or a number.
Example three: the segmentation condition determined according to the role and the content information comprises the following steps:
the type of the message with later sending time in the adjacent messages is a notice; optionally, the speaker of the notification may also be restricted to a particular role, such as a teacher, group administrator, and the like.
Example four: the segmentation condition determined according to the time, the role and the content information comprises the following steps:
the type of the message with the later sending time in the adjacent messages is a web page link or a number, and the sending time interval is more than a preset time length, such as 40 minutes. Alternatively, it can be defined that adjacent messages are sent from the same role, and that a message sent later in time in an adjacent message is closer to the next message than a message sent earlier in time in the adjacent message.
Optionally, the consistency condition may include one or more of various conditions in the following examples:
the message with later sending time in the adjacent messages is an empty message;
the message with later sending time consists of emoticons and long texts; wherein, the long text is a text with more words than a preset threshold value, such as 3, 4, etc.;
messages sent later in time consist of numbers;
messages that are sent later in time are composed of person names and non-text messages.
According to the segmentation condition and the consistency condition, whether the adjacent messages meeting the condition need to be segmented or not can be accurately judged.
In the embodiment of the application, whether the adjacent messages meet the segmentation condition or the consistency condition is determined according to the session segmentation model. And when the adjacent messages are determined not to meet the segmentation condition or the consistency condition, a session segmentation model is utilized to determine whether to segment the adjacent messages in time. Or after all the to-be-processed adjacent messages which do not meet the segmentation condition or the consistency condition in the target session are determined, determining whether to segment the adjacent messages or not by using the session segmentation model.
Therefore, the method can judge whether the adjacent messages in the target session need to be segmented or not based on the preset segmentation condition and the preset consistency condition. And under the condition that whether the adjacent messages need to be segmented cannot be determined based on the preset segmentation condition and the preset coherence condition, determining whether to segment the adjacent messages according to a session segmentation model. Therefore, the accuracy of segmenting the target session can be improved by setting the segmentation condition and the consistency condition, the misjudgment of adjacent messages which do not accord with the segmentation condition or the consistency condition is avoided by using the session segmentation model, and the accuracy of segmenting the target session is further improved.
As an exemplary embodiment, the session segmentation method may further include:
and training to obtain a session segmentation model by using the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session.
The above steps may be performed after determining that all of the pending neighbor messages in the target session do not meet the slicing condition nor the consistency condition.
Illustratively, the adjacent messages meeting the segmentation condition are marked as segmentation, the adjacent messages meeting the consistency condition are marked as non-segmentation, the adjacent messages are used as training data to train a conversation segmentation model, and based on the conversation segmentation model, a segmentation judgment result can be output according to the input adjacent messages.
The session segmentation model is obtained by training based on adjacent messages meeting the segmentation condition and the consistency condition in the target session, so that the context characteristics in the target session are obtained. The accuracy of the segmentation judgment result of other adjacent messages in the target session can be improved. Therefore, according to the exemplary embodiment, the accuracy of the method for performing segmentation judgment based on the preset conditions can be improved by setting strict preset conditions, the model is trained by using the accurate judgment result obtained based on the preset conditions, and the accuracy of the method for performing segmentation judgment based on the model is improved, so that the conversation can be accurately segmented by combining the two methods of judgment based on the preset conditions and judgment based on the model.
Illustratively, the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session may include:
acquiring an initial model obtained based on session scene corpus training;
and training the initial model by using the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session to obtain a session segmentation model.
For example, a session scene related to the target session, such as corpus under a session scene of an educational group, is used to obtain a pre-trained model, such as a BERT (Bidirectional Encoder representation from transducers) model.
According to the exemplary embodiment, an initial model with better prediction capability in a conversation scene is obtained by training based on a large amount of conversation scene linguistic data, and then the initial model is trained by utilizing adjacent messages which can obtain a segmentation judgment result based on a preset condition in a target conversation to obtain a conversation segmentation model, so that the overall training data is increased, and context information is considered more, so that the model has good performance in the target conversation.
Optionally, the initial model is a BERT model, and the BERT model can convert static embedding in a word into dynamic, so that a word ambiguity problem can be solved; moreover, the BERT model uses deep networks, which results in more abstract context-dependent features at a high level. Therefore, adopting the BERT model as the initial model can further improve the effect of the session segmentation model.
As an exemplary embodiment, the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the coherence condition in the target session includes:
connecting adjacent messages meeting the segmentation conditions in the target session with at least one message in front of the adjacent messages to obtain first training data marked as segmentation;
connecting the adjacent message meeting the consistency condition in the target session with at least one message in front of the adjacent message to obtain second training data marked as non-segmentation;
the session segmentation model is trained using the first training data and the second training data.
The number of at least one message before the adjacent message may be a preset number, or may be a number from a last slicing position to a message between adjacent messages.
For example, if the adjacent messages meeting the slicing condition include the 4 th and 5 th messages in the conversation, and the preset number is 2, the 2 messages before the 4 th message are connected with the 4 th and 5 th messages, that is, the 2 nd to 5 th messages are connected and marked as slicing. If consecutive eligible adjacent messages include the 5 th and 6 th messages in the conversation, the 3 rd to 6 th messages are concatenated and marked as not fragmented. Alternatively, a binary number is used as the label information, for example, 0 is used to indicate the slicing, and 1 is used to indicate the non-slicing.
With the above exemplary embodiment, the format of the training data may refer to the following example:
In the above format, the eligible neighbor messages are message 3 and message 4, and the split position is identified by comma between message 3 and message 4.
According to the exemplary embodiment, the session segmentation model can be trained by combining more context information, so that the effect of judgment segmentation of the session segmentation model is further improved.
In some embodiments, only adjacent messages may be concatenated and labeled, and the format of the training data may refer to the following examples:
message 3, message 4, annotation information.
As an exemplary embodiment, referring to fig. 2, the session segmentation method may further include:
step S15, obtaining the predicted segmentation position in the target session according to the segmentation result of the adjacent message;
step S16, calculating to obtain a first evaluation index according to the predicted segmentation position and the pre-marked actual segmentation position; wherein the first evaluation index is used for adjusting at least one of a segmentation condition, a coherence condition and a session segmentation model.
And the predicted segmentation position represents the segmentation position of the target session predicted by using the preset condition and the session segmentation model. For example, if it is determined to slice a 3 rd message and a 4 th message, the predicted slice location is between the 3 rd and the 4 th. After the segmentation result is respectively determined for each to-be-processed adjacent message of the target session, all the predicted segmentation positions in the target session can be obtained.
The actual segmentation position may be a manually marked segmentation position, for example, a segmentation position obtained by segmenting the context in the case that it is manually determined that the contexts do not belong to the same topic.
The first evaluation index may include a first accuracy and/or a first recall.
The calculation formula of the first accuracy may refer to the following example:
Wherein the content of the first and second substances,
for the number of sessions used for calculating the first evaluation index, e.g. from the predicted slicing positions and the pre-labeled actual slicing positions of the target session and the other 3 sessions, a first accuracy is calculated, then d = 4.
For the same number of predicted slicing positions as the actual slicing positions in the ith session,
the number of slice locations is sliced for the prediction of the ith session. Namely, it is
Indicating how many of the predicted slicing positions in the ith session are quasiThe exact, i.e. slicing accuracy of the ith session. First degree of accuracy
Represents the average of the slicing accuracy of a plurality of sessions.
The calculation formula of the first recall rate may refer to the following example:
Wherein the content of the first and second substances,
is the number of sessions used to calculate the first rating index.
For the same number of predicted slicing positions as the actual slicing positions in the ith session,
the number of actual slicing positions for the ith session. Namely, it is
Indicating how many of the actual slicing positions in the ith session are accurately predicted, i.e., the predicted recall rate of the ith session. First recall rate
Represents the average of the predicted recall rates for a plurality of sessions.
Since the segmentation position is the basis for actual segmentation, the exemplary embodiment adjusts at least one of the segmentation condition, the continuity condition and the session segmentation model by using the first evaluation index calculated by the segmentation position, so that the prediction accuracy of the segmentation position can be improved.
As an exemplary embodiment, referring to fig. 3, the session segmentation method may further include:
step S17, according to the prediction segmentation position, segmenting the target session into a plurality of prediction session segments;
step S18, calculating a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; and the second evaluation index is used for evaluating the segmentation quality of the target session.
For example, the target session may be sliced into 4 predicted session segments based on 3 predicted slice positions. Correspondingly, the actual session segment may be a session segment obtained by segmenting the target session according to the manually marked actual segmentation position.
The second evaluation index may include a second accuracy and/or a second recall.
The calculation formula of the second accuracy may refer to the following example:
Wherein the content of the first and second substances,
in order to predict the same number of session segments as the actual session segments, the predicted session segments are considered to be the same as the actual session segments when the starting points and the ending points of the predicted session segments and the actual session segments are identical.
To predict the number of session segments, the second degree of accuracy indicates how accurate the predicted session segments are in the target session.
The calculation formula of the second recall rate may refer to the following example:
Wherein the content of the first and second substances,
to predict the same number of session segments as the actual session segments,
the second recall rate represents how many actual session segments in the target session are accurately predicted for the number of actual session segments.
The accurate segmentation can be represented only when the session segment is accurate, namely the segmentation positions of the session segment before and after are accurate. Thus, the session segment accuracy is the optimization goal for slicing the target session. According to the exemplary embodiment, the second evaluation index obtained by calculating the session segment is used for evaluating the segmentation quality of the target session, so that the segmentation method can be improved towards a correct optimization target, and the segmentation quality is improved.
Therefore, in the embodiment of the application, two evaluation standards are provided on the final segmentation effect evaluation: the accuracy and recall of each session segment and the accuracy and recall of each segmentation position. The accuracy and recall rate of each session section are final optimization targets; the accuracy and the recall rate of the segmentation positions can visually check the related information of the segmentation positions, the final segmentation position number statistics cannot be influenced by the wrong segmentation positions in some segments, the segmentation positions are used for adjusting the preset conditions or models, and the efficiency is high.
Fig. 4 shows a schematic diagram of an application example of the present application. As shown in fig. 4, in this application example, the target session is a session in an educational scenario, and the target session includes information such as student ID (Identity), teacher ID, text content (i.e., message), role, and time. First, content type recognition is performed on the target session, recognizing that each message includes a link, a notification, an emoticon, a picture, a start word, an end word, a number, a letter, null data, or other types of messages. And then, segmenting the session according to preset conditions and one or more of time, role and content information. And if the segmentation cannot be determined according to the preset conditions, determining whether the segmentation is performed by using a semantic model, such as a deep neural network model. And the segmentation of the session is realized by combining the model and the preset condition. Finally, a plurality of predicted session segments in the target session are output, each predicted session segment including information of student ID, teacher ID and text content, role, time, session segment ID (sess _ ID1, sess _ ID2, sess _ idm), and the like.
Fig. 5 shows a schematic diagram of another application example of the present application. As shown in fig. 5, in this application example, a target session is first read, and the target session includes n messages. Then, the target session is segmented according to the information (time, content, role) of the three dimensions and preset conditions. Inputting the ith message into a BERT model, and judging whether the ith message and the former are segmented or not. And if the segmentation is carried out according to the preset condition, enabling i = i +1, re-inputting the ith message into the BERT model and judging whether the segmentation is carried out or not. If not, connecting the data from the last segmentation position to the ith message according to the format of 'the preceding, the ith message' and inputting the data into a BERT model for prediction. And if i is less than n, enabling i = i +1, returning the step of inputting the ith message to the BERT model until i = n, and outputting the segmentation result of the target session.
In summary, the method of the embodiment of the present application may determine whether the adjacent messages in the target session need to be segmented based on the preset segmentation condition and the preset coherence condition. And under the condition that whether the adjacent messages need to be segmented cannot be determined based on the preset segmentation condition and the preset coherence condition, determining whether to segment the adjacent messages according to a session segmentation model. Therefore, the accuracy of segmenting the target session can be improved by setting the segmentation condition and the consistency condition, the misjudgment of adjacent messages which do not accord with the segmentation condition or the consistency condition is avoided by using the session segmentation model, and the accuracy of segmenting the target session is further improved.
Fig. 6 is a block diagram illustrating a structure of a session segmentation apparatus according to an embodiment of the present invention. As shown in fig. 6, the apparatus may include:
an obtaining module 610, configured to obtain an adjacent message to be processed in a target session;
the segmentation module 620 is configured to determine to segment the adjacent messages when the adjacent messages meet a preset segmentation condition;
a consistency module 630, configured to determine not to segment the adjacent messages when the adjacent messages meet a preset consistency condition;
the determining module 640 is configured to determine whether to segment the adjacent messages according to the session segmentation model when the adjacent messages do not meet the segmentation condition and do not meet the consistency condition.
Exemplarily, referring to fig. 7, the apparatus further comprises:
the training module 710 is configured to train to obtain a session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session.
Illustratively, referring to fig. 7, the training module 710 includes:
an initialization unit 711, configured to obtain an initial model obtained through corpus training based on a session scene;
the first training unit 712 is configured to train the initial model by using the neighboring message meeting the segmentation condition and the neighboring message meeting the consistency condition in the target session, so as to obtain a session segmentation model.
Illustratively, referring to fig. 7, a training module, 710, includes:
a first connection unit 713, configured to connect an adjacent message that meets the segmentation condition in the target session with at least one message that precedes the adjacent message, to obtain first training data labeled as segmentation;
a second connecting unit 714, configured to connect an adjacent message that meets a consistency condition in the target session with at least one message that precedes the adjacent message, to obtain second training data labeled as non-segmented;
a second training unit 715, configured to train the session segmentation model using the first training data and the second training data.
Illustratively, the slicing conditions include at least one of the following conditions:
the speakers of adjacent messages belong to the same role, and adjacent messages are sent out on different dates;
the sending time interval of the adjacent messages is greater than the preset time length, and the message with later sending time in the adjacent messages comprises a preset starting language;
sending out adjacent messages on different dates, wherein the messages with earlier sending time in the adjacent messages comprise preset end words;
the adjacent messages are sent out on different dates, and the type of the message with the earlier sending time in the adjacent messages is a webpage link;
the type of the message with later sending time in the adjacent messages is a notice;
the type of the message with the later sending time in the adjacent messages is a webpage link or a number, and the sending time interval of the adjacent messages is more than the preset time length.
Illustratively, the consistency condition includes at least one of the following conditions:
the message with later sending time in the adjacent messages is an empty message;
the message with later sending time consists of emoticons and long texts; the long text is a text with more words than a preset threshold value;
messages sent later in time consist of numbers;
messages that are sent later in time are composed of person names and non-text messages.
Exemplarily, referring to fig. 7, the apparatus further comprises:
a segmentation position module 720, configured to obtain a predicted segmentation position in the target session according to the segmentation result of the adjacent message;
the first evaluation module 730 is configured to calculate a first evaluation index according to the predicted segmentation position and a pre-labeled actual segmentation position; wherein the first evaluation index is used for adjusting at least one of a segmentation condition, a coherence condition and a session segmentation model.
Exemplarily, referring to fig. 7, the apparatus further comprises:
a session segment module 740, configured to segment the target session into a plurality of predicted session segments according to the predicted segmentation position;
the second evaluation module 750 is configured to calculate a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; and the second evaluation index is used for evaluating the segmentation quality of the target session.
The functions of each module in each apparatus in the embodiments of the present invention may refer to the corresponding description in the above method, and are not described herein again.
Fig. 8 shows a block diagram of an electronic device according to an embodiment of the present application. As shown in fig. 8, the electronic apparatus includes: a memory 910 and a processor 920, the memory 910 having stored therein computer programs operable on the processor 920. The processor 920, when executing the computer program, implements the methods in the embodiments described above. The number of the memory 910 and the processor 920 may be one or more.
The electronic device further includes:
and a communication interface 930 for communicating with an external device to perform data interactive transmission.
If the memory 910, the processor 920 and the communication interface 930 are implemented independently, the memory 910, the processor 920 and the communication interface 930 may be connected to each other through a bus and perform communication with each other. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended ISA (enhanced Industry Standard Architecture) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 8, but this is not intended to represent only one bus or type of bus.
Optionally, in an implementation, if the memory 910, the processor 920 and the communication interface 930 are integrated on a chip, the memory 910, the processor 920 and the communication interface 930 may complete communication with each other through an internal interface.
Embodiments of the present invention provide a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, the computer program implements the method provided in the embodiments of the present application.
The embodiment of the present application further provides a chip, where the chip includes a processor, and is configured to call and execute the instruction stored in the memory from the memory, so that the communication device in which the chip is installed executes the method provided in the embodiment of the present application.
An embodiment of the present application further provides a chip, including: the system comprises an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected through an internal connection path, the processor is used for executing codes in the memory, and when the codes are executed, the processor is used for executing the method provided by the embodiment of the application.
It should be understood that the processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. A general purpose processor may be a microprocessor or any conventional processor or the like. It is noted that the processor may be an advanced reduced instruction set machine (ARM) architecture supported processor.
Further, optionally, the memory may include a read-only memory and a random access memory, and may further include a nonvolatile random access memory. The memory may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The non-volatile memory may include a read-only memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an electrically Erasable EPROM (EEPROM), or a flash memory. Volatile memory can include Random Access Memory (RAM), which acts as external cache memory. By way of example, and not limitation, many forms of RAM are available. For example, Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), double data rate synchronous SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM), and direct memory bus RAM (DR RAM).
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The procedures or functions according to the present application are generated in whole or in part when the computer program instructions are loaded and executed on a computer. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present application, "a plurality" means two or more unless specifically limited otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process. And the scope of the preferred embodiments of the present application includes other implementations in which functions may be performed out of the order shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions.
It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the method of the above embodiments may be implemented by hardware that is configured to be instructed to perform the relevant steps by a program, which may be stored in a computer-readable storage medium, and which, when executed, includes one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present application may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module may also be stored in a computer-readable storage medium if it is implemented in the form of a software functional module and sold or used as a separate product. The storage medium may be a read-only memory, a magnetic or optical disk, or the like.
While the present invention has been described with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (18)
1. A session segmentation method is characterized by comprising the following steps:
acquiring adjacent messages to be processed in a target session;
under the condition that the adjacent messages accord with preset segmentation conditions, determining to segment the adjacent messages;
under the condition that the adjacent messages meet preset consistency conditions, the adjacent messages are determined not to be segmented;
and under the condition that the adjacent messages do not accord with the segmentation conditions and the consistency conditions, determining whether to segment the adjacent messages according to a session segmentation model.
2. The method of claim 1, further comprising:
and training to obtain the session segmentation model by utilizing the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session.
3. The method according to claim 2, wherein the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session comprises:
acquiring an initial model obtained based on session scene corpus training;
and training the initial model by using the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target conversation to obtain the conversation segmentation model.
4. The method according to claim 2, wherein the training of the session segmentation model by using the neighboring messages meeting the segmentation condition and the neighboring messages meeting the consistency condition in the target session comprises:
connecting the adjacent message meeting the segmentation condition in the target session with at least one message before the adjacent message to obtain first training data marked as segmentation;
connecting the adjacent message meeting the consistency condition in the target session with at least one message before the adjacent message to obtain second training data marked as non-segmentation;
training the session segmentation model using the first training data and the second training data.
5. The method of claim 1, wherein the slicing conditions comprise at least one of the following conditions:
the speakers of the adjacent messages belong to the same role, and the adjacent messages are sent out on different dates;
the sending time interval of the adjacent messages is greater than a preset time length, and the message with later sending time in the adjacent messages comprises a preset starting language;
the adjacent messages are sent out on different dates, and the messages with earlier sending time in the adjacent messages comprise preset end words;
the adjacent messages are sent out on different dates, and the type of the message with the earlier sending time in the adjacent messages is a webpage link;
the type of the message with later sending time in the adjacent messages is a notice;
the type of the message with the later sending time in the adjacent messages is a webpage link or a number, and the sending time interval of the adjacent messages is larger than the preset time length.
6. The method of claim 1, wherein the consistent condition comprises at least one of the following conditions:
the message with later sending time in the adjacent messages is an empty message;
the message with the later sending time consists of emoticons and long texts; the long text is a text with more words than a preset threshold value;
the message with the later sending time consists of numbers;
the message sent later in time consists of a person name and a non-text message.
7. The method of any one of claims 1 to 6, further comprising:
obtaining a predicted segmentation position in the target session according to the segmentation result of the adjacent message;
calculating to obtain a first evaluation index according to the predicted segmentation position and a pre-marked actual segmentation position; wherein the first evaluation index is used for adjusting at least one of the segmentation condition, the consistency condition and the conversation segmentation model.
8. The method of claim 7, further comprising:
according to the prediction segmentation position, segmenting the target session into a plurality of prediction session segments;
calculating to obtain a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; wherein the second evaluation index is used for evaluating the segmentation quality of the target session.
9. A session splitting apparatus, comprising:
the acquisition module is used for acquiring adjacent messages to be processed in the target session;
the segmentation module is used for determining to segment the adjacent messages under the condition that the adjacent messages meet the preset segmentation conditions;
the consistency module is used for determining not to segment the adjacent messages under the condition that the adjacent messages meet preset consistency conditions;
and the determining module is used for determining whether to segment the adjacent messages according to a session segmentation model under the condition that the adjacent messages do not accord with the segmentation conditions and the consistency conditions.
10. The apparatus of claim 9, further comprising:
and the training module is used for training to obtain the session segmentation model by utilizing the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target session.
11. The apparatus of claim 10, wherein the training module comprises:
the initialization unit is used for acquiring an initial model obtained based on session scene corpus training;
and the first training unit is used for training the initial model by utilizing the adjacent messages meeting the segmentation condition and the adjacent messages meeting the consistency condition in the target conversation to obtain the conversation segmentation model.
12. The apparatus of claim 10, wherein the training module comprises:
the first connecting unit is used for connecting the adjacent message meeting the segmentation condition in the target session with at least one message before the adjacent message to obtain first training data marked as segmentation;
a second connection unit, configured to connect an adjacent message that meets the coherence condition in the target session with at least one message before the adjacent message, to obtain second training data labeled as non-segmented;
and the second training unit is used for training the conversation segmentation model by utilizing the first training data and the second training data.
13. The apparatus of claim 9, wherein the slicing conditions comprise at least one of the following conditions:
the speakers of the adjacent messages belong to the same role, and the adjacent messages are sent out on different dates;
the sending time interval of the adjacent messages is greater than a preset time length, and the message with later sending time in the adjacent messages comprises a preset starting language;
the adjacent messages are sent out on different dates, and the messages with earlier sending time in the adjacent messages comprise preset end words;
the adjacent messages are sent out on different dates, and the type of the message with the earlier sending time in the adjacent messages is a webpage link;
the type of the message with later sending time in the adjacent messages is a notice;
the type of the message with the later sending time in the adjacent messages is a webpage link or a number, and the sending time interval of the adjacent messages is larger than the preset time length.
14. The apparatus of claim 9, wherein the consistent condition comprises at least one of:
the message with later sending time in the adjacent messages is an empty message;
the message with the later sending time consists of emoticons and long texts; the long text is a text with more words than a preset threshold value;
the message with the later sending time consists of numbers;
the message sent later in time consists of a person name and a non-text message.
15. The apparatus of any one of claims 9 to 14, further comprising:
the segmentation position module is used for obtaining a prediction segmentation position in the target session according to the segmentation result of the adjacent message;
the first evaluation module is used for calculating to obtain a first evaluation index according to the predicted segmentation position and a pre-marked actual segmentation position; wherein the first evaluation index is used for adjusting at least one of the segmentation condition, the consistency condition and the conversation segmentation model.
16. The apparatus of claim 15, further comprising:
the session segment module is used for segmenting the target session into a plurality of predicted session segments according to the predicted segmentation positions;
the second evaluation module is used for calculating to obtain a second evaluation index according to the plurality of predicted session segments and the pre-labeled actual session segments; wherein the second evaluation index is used for evaluating the segmentation quality of the target session.
17. An electronic device, comprising: a processor and a memory, the memory having stored therein instructions that are loaded and executed by the processor to implement the method of any of claims 1 to 8.
18. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010853510.9A CN111708866B (en) | 2020-08-24 | 2020-08-24 | Session segmentation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010853510.9A CN111708866B (en) | 2020-08-24 | 2020-08-24 | Session segmentation method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111708866A true CN111708866A (en) | 2020-09-25 |
| CN111708866B CN111708866B (en) | 2020-12-11 |
Family
ID=72547391
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010853510.9A Active CN111708866B (en) | 2020-08-24 | 2020-08-24 | Session segmentation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111708866B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111970311A (en) * | 2020-10-23 | 2020-11-20 | 北京世纪好未来教育科技有限公司 | Session segmentation method, electronic device and computer readable medium |
| CN112203094A (en) * | 2020-11-16 | 2021-01-08 | 北京世纪好未来教育科技有限公司 | Encoding method, encoding device, electronic device and storage medium |
| CN112437125A (en) * | 2020-11-10 | 2021-03-02 | 北京百度网讯科技有限公司 | Information concurrent processing method and device, electronic equipment and storage medium |
| CN113569017A (en) * | 2021-01-28 | 2021-10-29 | 腾讯科技（深圳）有限公司 | Model processing method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101119326A (en) * | 2006-08-04 | 2008-02-06 | 腾讯科技（深圳）有限公司 | Method and device for managing instant communication conversation recording |
| CN107066450A (en) * | 2017-05-27 | 2017-08-18 | 国家计算机网络与信息安全管理中心 | A kind of instant communication session segmentation technique and method based on study |
| US20190385739A1 (en) * | 2017-06-28 | 2019-12-19 | Boe Technology Group Co., Ltd. | Triage dialogue method, device and system |
| CN110691025A (en) * | 2018-07-06 | 2020-01-14 | 阿里巴巴集团控股有限公司 | Message segmentation method and device |
| CN111125317A (en) * | 2019-12-27 | 2020-05-08 | 携程计算机技术（上海）有限公司 | Model training, classification, system, device and medium for conversational text classification |
-
2020
- 2020-08-24 CN CN202010853510.9A patent/CN111708866B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101119326A (en) * | 2006-08-04 | 2008-02-06 | 腾讯科技（深圳）有限公司 | Method and device for managing instant communication conversation recording |
| CN107066450A (en) * | 2017-05-27 | 2017-08-18 | 国家计算机网络与信息安全管理中心 | A kind of instant communication session segmentation technique and method based on study |
| US20190385739A1 (en) * | 2017-06-28 | 2019-12-19 | Boe Technology Group Co., Ltd. | Triage dialogue method, device and system |
| CN110691025A (en) * | 2018-07-06 | 2020-01-14 | 阿里巴巴集团控股有限公司 | Message segmentation method and device |
| CN111125317A (en) * | 2019-12-27 | 2020-05-08 | 携程计算机技术（上海）有限公司 | Model training, classification, system, device and medium for conversational text classification |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111970311A (en) * | 2020-10-23 | 2020-11-20 | 北京世纪好未来教育科技有限公司 | Session segmentation method, electronic device and computer readable medium |
| CN112437125A (en) * | 2020-11-10 | 2021-03-02 | 北京百度网讯科技有限公司 | Information concurrent processing method and device, electronic equipment and storage medium |
| CN112203094A (en) * | 2020-11-16 | 2021-01-08 | 北京世纪好未来教育科技有限公司 | Encoding method, encoding device, electronic device and storage medium |
| CN112203094B (en) * | 2020-11-16 | 2024-05-14 | 北京世纪好未来教育科技有限公司 | Encoding method, encoding device, electronic equipment and storage medium |
| CN113569017A (en) * | 2021-01-28 | 2021-10-29 | 腾讯科技（深圳）有限公司 | Model processing method and device, electronic equipment and storage medium |
| CN113569017B (en) * | 2021-01-28 | 2024-05-10 | 腾讯科技（深圳）有限公司 | Model processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111708866B (en) | 2020-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111708866B (en) | Session segmentation method and device, electronic equipment and storage medium | |
| US20200042613A1 (en) | Processing an incomplete message with a neural network to generate suggested messages | |
| CN112365894B (en) | AI-based composite voice interaction method and device and computer equipment | |
| CN108682420B (en) | Audio and video call dialect recognition method and terminal equipment | |
| JP6677419B2 (en) | Voice interaction method and apparatus | |
| CN110493019B (en) | Automatic generation method, device, equipment and storage medium of conference summary | |
| CN109074547B (en) | Text message ordering based on message content | |
| CN108959388B (en) | Information generation method and device | |
| RU2670029C2 (en) | System and method of automatic message moderation | |
| EP3614297A1 (en) | Hybrid natural language understanding | |
| CN111079410A (en) | Text recognition method and device, electronic equipment and storage medium | |
| CN111079433B (en) | Event extraction method and device and electronic equipment | |
| CN114756659A (en) | Language model training method, device, equipment and storage medium | |
| WO2018075190A1 (en) | Systems and methods for handling formality in translations of text | |
| EP3796191A1 (en) | Chapter-level text translation method and device | |
| US20200142996A1 (en) | Method, apparatus and computer program product for improving deep question-answering (qa) applications using feedback from retrieval qa applications | |
| CN113255368B (en) | Method and device for emotion analysis of text data and related equipment | |
| CN115422936A (en) | Entity identification method, entity identification device, computer equipment and storage medium | |
| EP2846499B1 (en) | Method And Device For Classifying A Message | |
| CN114141235A (en) | Voice corpus generation method and device, computer equipment and storage medium | |
| CN112840627B (en) | Information processing method, related device and computer readable storage medium | |
| CN114846543A (en) | Voice recognition result detection method and device and storage medium | |
| CN111552785A (en) | Method and device for updating database of human-computer interaction system, computer equipment and medium | |
| CN113158692B (en) | Semantic recognition-based multi-intention processing method, system, equipment and storage medium | |
| US11941358B2 (en) | Identifying messages entered into an incorrect conversation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |