CN115757551B - Event key information mining and predicting method - Google Patents
Event key information mining and predicting method Download PDFInfo
- Publication number
- CN115757551B CN115757551B CN202211520106.5A CN202211520106A CN115757551B CN 115757551 B CN115757551 B CN 115757551B CN 202211520106 A CN202211520106 A CN 202211520106A CN 115757551 B CN115757551 B CN 115757551B
- Authority
- CN
- China
- Prior art keywords
- event
- data
- user
- sample
- result
- 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.)
- Active
Links
Abstract
The application discloses a method for mining and predicting event key information, which comprises the following steps: constructing a model foundation; predictive model training, comprising: invoking past event data of a certain athlete as a history sample, randomly sampling and dividing the history sample to obtain a test sample and a training sample; training and testing a preset prediction model algorithm based on the test sample and the training sample, and outputting and storing the preset prediction model algorithm; event keypoint mining, comprising: acquiring video data of a current event, and performing face recognition on a top player to obtain identity information of the top player; calling a matched prediction model according to the identity information of the player in the field; and acquiring event data of the current event, extracting characteristic variables, importing a matched prediction model for analysis to obtain a prediction result, and outputting the prediction result serving as a key prompt. The application has the advantage of reducing the working pressure and difficulty of related personnel for broadcasting and explaining sports events.
Description
Technical Field
The application relates to the technical field of big sports data, in particular to a method for mining and predicting key information of an event.
Background
Today, where information technology is rapidly evolving, data information is an important resource, and technological revolution derived from big data is affecting each industry year by year.
Taking sports big data as an example: traditionally, for athlete training, the self-quality of each coach is relied on; most of them are: the training performance of the athlete is observed by a coach, and a training task is formulated by combining the past experience of the coach, so that the athlete reaches the optimal state before the athlete plays.
At present, along with the introduction of big data concepts in sports, various physiological parameter changes of athletes are monitored in real time and for a long time by means of intelligent wearing equipment, and a big data system for guiding a coach to adjust the intensity and quantity of training tasks is combined with intelligent analysis of past training records so as to realize more scientific, accurate and effective training.
Based on the above, the inventors considered that: at present, the real-time broadcasting and explanation of the sports event is similar to the training task formulated by the coach, and the working experience and the professional literacy of related personnel are excessively depended, so that the working pressure and the difficulty of the practitioner are high, and the application provides a new technical scheme.
Disclosure of Invention
In order to reduce the work difficulty of broadcasting and explaining a sports event, the application provides a method for mining and predicting key information of the event.
The application provides a method for mining and predicting key information of an event, which adopts the following technical scheme:
a method for mining and predicting event key information comprises the following steps:
q1, constructing a model foundation, which comprises the following steps:
acquiring multi-field event information as sample data;
analyzing sample data based on pre-recorded event score rules and score factors, and defining characteristic variables required by predictive analysis;
feature extraction is carried out on the analyzed sample data, and a mapping relation between feature variable combination and a scoring result is established based on sequential logic;
defining feature variable combination rules based on the mapping relation and the pre-recorded feature combination examples;
q2, predictive model training, comprising:
collecting and calling past event data of any athlete as a history sample;
extracting features of the historical samples to generate athlete personal specification data;
randomly sampling and dividing the historical samples to obtain a test sample and a training sample;
training and testing a preset prediction model algorithm based on the test sample and the training sample, outputting the preset prediction model algorithm, and storing the preset prediction model algorithm according to the identity of the athlete;
q3, mining event key points, which comprises:
acquiring video data of a current event, and performing face recognition on a top player to obtain identity information of the top player;
according to the identity information, calling a matched prediction model;
and acquiring event data of the current event, extracting characteristic variables, importing a matched prediction model for analysis to obtain a prediction result, and outputting the prediction result serving as a key prompt.
Alternatively to this, the method may comprise,
defining a preamble characteristic variable corresponding to the primary score and combining the characteristic variables into a group of data;
counting each group of data of the same athlete to obtain the same characteristic value and the frequency of occurrence of the characteristic variable combination as the ratio d, and defining the weight value of the characteristic variable based on the frequency of occurrence as the ratio d;
respectively calculating standard deviation/similarity of each characteristic variable;
judging whether the standard deviation/similarity is smaller than a preset first threshold value, if not, calculating the duty ratios of a plurality of different preset value ranges of the same characteristic variable, and recording the duty ratios as a single duty ratio dm;
and synthesizing the weight value and the single-quantity duty ratio dm, obtaining a new weight value and outputting the new weight value.
Optionally, the method further comprises:
acquiring user speaking data of a pre-docked event online platform;
keyword/word recognition and positioning are carried out on the speaking data of the user;
based on keyword/word, grasping prejudgement behavior information in user speaking, and based on time sequence logic, comparing score data of subsequent event;
counting the correct times of the prejudging actions of the user, and defining the user with the times exceeding a preset first threshold as a reference user;
establishing a user file for a reference user, and recording prejudging behavior information, corresponding event true score data and matched athletes of the prejudging behavior by the user file;
making reference user prejudgement statistics for the present athlete of the current event, and outputting the statistical result as key prompt.
Optionally, the positioning includes:
carrying out semantic analysis on a certain keyword/word, speaking front and back, judging whether to trigger the pre-judgment of the player's behavior, if so, positioning successfully, and executing the next step; if not, keyword/word recognition is returned.
Optionally, the method further comprises:
counting the prejudgment result of the reference user and marking the prejudgment result as a data set A;
comparing the data set A with corresponding event score data, and judging consistency, wherein the result is marked as Km;
comparing the model prediction result with corresponding event score data, and judging consistency, wherein the result is recorded as Kn;
if the number of the consistent times in the historical Km data set is larger than the number of the consistent times in the Kn data set, defining the user pre-judgment as a characteristic variable, and updating the prediction.
Optionally, the comparing the data set a with the corresponding event score data, and performing consistency determination includes:
counting the number of various user prejudging results, and comparing the user prejudging result with the largest number with event score data; and if the same, the same is judged as the same.
Optionally, capturing preference trend information in the speech of the reference user based on the keyword/word pairs, and identifying the category of the preference trend information; wherein the categories include forward preferences and negative preferences;
matching preference trend information with a corresponding athlete based on a reference user utterance;
when counting data set a, reference users that are negative preferences for the current athlete and whose frequency exceeds a second threshold are skipped.
Optionally, the method further comprises: and judging the following behavior of the reference user speaking based on the keywords/words, and skipping the reference user when the following behavior is judged to be following the wind, and counting the data set A.
In summary, the present application includes at least one of the following beneficial technical effects: the event broadcasting and explaining personnel can predict the subsequent performance of a certain athlete in the current event by using the method, take the prediction result as guidance, pay attention to, explain and analyze the behaviors of the corresponding athlete, and the like, so as to realize the excavation and the early discovery of the key information of the event, improve the effect of the event broadcasting and explaining, and reduce the related working pressure and difficulty.
Drawings
FIG. 1 is a schematic illustration of the main process of the present application.
Detailed Description
The present application will be described in further detail with reference to fig. 1.
The embodiment of the application discloses a method for mining and predicting event key information, which can be realized by loading and running a processor through a corresponding computer program.
Referring to fig. 1, the event key information mining and prediction method includes:
q1, constructing a model foundation, which comprises the following steps:
acquiring multi-field competition information as sample data;
analyzing sample data based on pre-recorded event score rules and score factors, and defining characteristic variables required by predictive analysis.
The above-described parsed sample data refers to, for example: and when the event information is voice broadcasting content, performing voice recognition on the audio file, converting the text into a text, performing key word and sentence extraction on the text, and extracting a predefined event characteristic sentence. Wherein, the event feature sentence refers to: the real-time event data excluding the explanation of the event, the resolution of the person, the advertisement popularization and the out-of-field information can be player behaviors.
And the definition of the characteristic variables can be matched and screened according to the pre-uploaded event scoring rules and scoring factors.
Subsequently, carrying out feature extraction on the analyzed sample data, and establishing a mapping relation between feature variable combination and a scoring result based on sequential logic;
based on the mapping relation and the pre-entered feature combination example, feature variable combination rules are defined.
In this embodiment, taking football as an example, the feature variables may be: location, number of interceptors, type of goal, time of goal, distance of goal, etc.
Regarding feature variable combinations, for example:
combination a, which is sufficiently comprehensive for event data: coordinates (X, Y) position + intercept (0) +departure time (34 min) +goal distance (10 m);
for limited combinations of event data B: coordinates (X, Y) position + intercept (2 people) +open time (34 min);
for combination C without balls: intercept (0) +coordinate (X, Y) position.
Q2, predictive model training, comprising:
collecting and calling past event data of any athlete as a history sample;
and extracting the characteristics of the historical samples to generate athlete personal specification data.
In one embodiment of the present application, the data preprocessing step is determined according to the following prediction model, for example: when the integrated learning models such as logistic regression, decision trees, boosting and Bagging based on the decision trees are selected, the data standardization processing is not needed in the process in consideration of insensitivity to the feature values. The difference of two characteristic variables with the characteristic value being understandable can be processed by variance/standard deviation when the difference is large and the prediction model is sensitive to the difference. According to the above-mentioned available class table data, for example:
column_list= [ 'position', 'interceptor', 'time to open', 'range' ];
formath_methor in[……]；
temp=data.groupby ('ID', as_index=false) [ columns_list ]. Agg (math_method); the// ID, namely the identity of the athlete corresponding to the data;
temp.columns = ['ID'] + [name + x + math_methor for x in temp.columns[1:]]；
feature = pd.merge(feature,temp,how='left')。
after the required data are obtained, the following steps are carried out:
randomly sampling and dividing the historical samples to obtain a test sample and a training sample; such as: 15% as test, 75% as training;
defining model parameters; such as: the model relates to a decision tree, and the number of times of cross verification is defined;
and (5) invoking the training sample to train and predict a preset prediction model algorithm, and generating a prediction model. It will be appreciated that the predictive model algorithm described above is a background free choice, such as: random forest regression algorithm (rfr =random forest regressive ()), logistic regression algorithm.
After the model is initially trained:
calling a test sample to predict by a prediction model, generating a prediction result, and recording;
then, comparing the test sample with the prediction result to generate model evaluation data;
judging whether the evaluation passes or not, if so, outputting and storing the model; if not, the test sample and the training sample are regenerated, the training is performed again, and the test model is tested.
Q3, mining event key points, which comprises:
acquiring video data of a current event, and performing face recognition (searching a preset face database) on a live athlete to obtain identity information of the live athlete;
invoking a matched predictive model based on the presence athlete identity information;
and acquiring event data of the current event, extracting characteristic variables, importing a matched prediction model for analysis to obtain a prediction result, and outputting the prediction result serving as a key information prompt.
According to the content, the event broadcasting and explaining personnel can predict the subsequent performance of a certain athlete in the current event by using the method, take the prediction result as guidance, pay attention to, explain and analyze the behaviors of the corresponding athlete and the like, so that the key information of the event is excavated and found in advance, the effect of the event broadcasting and explaining is improved, and meanwhile, the related working pressure and difficulty are reduced.
In one embodiment of the method, the following is done for the feature variables:
defining a preamble characteristic variable corresponding to the primary score and combining the characteristic variables into a group of data;
counting each group of data of the same athlete to obtain the same characteristic value and the frequency of occurrence of the characteristic variable combination as the ratio d, and defining the weight value of the characteristic variable based on the frequency of occurrence as the ratio d;
respectively calculating standard deviation/similarity of each characteristic variable;
judging whether the standard deviation/similarity is smaller than a preset first threshold value, if not, calculating the duty ratios of a plurality of different preset value ranges of the same characteristic variable, and recording the duty ratios as a single duty ratio dm;
and (3) synthesizing the weight value and the single-quantity duty ratio dm (such as multiplication), obtaining a new weight value and outputting the new weight value.
It can be understood that in the prediction behavior, whether a certain result appears is judged, which is determined by the sum of the scores (contribution degrees) of the characteristic variables, and the score is high, which indicates that the corresponding result appears more probably; such as logistic regression as described above, the parameters of which relate to typed weight parameters; XGBoost (gradient lifting tree), the parameters of which relate to sample weights.
That is, the above-described processing steps of the feature variables of the present application may be used to define, update, and screen the feature variables by no longer classifying, such as by generating a feature thermodynamic diagram; also because, at this time, a single amount of the duty dm is introduced; namely, when the value of a certain characteristic variable is changed greatly, the influence on the result when the value of the certain characteristic variable is different is further refined, so that the result is more real.
For example: the difference in the influence of the coordinate (x=2, y=2) position and the coordinate (20, 20) position on the result is considered to be large, and thus a single amount of the duty dm is introduced.
It should be noted that if the defined model algorithm does not involve the weight concept, the weight calculation process is independent, the model prediction process is parallel, and the direct output result is used as another hint.
In one embodiment of the method, the method further comprises:
acquiring user speaking data of a pre-docked event online platform (an image-text live broadcast platform and a video live broadcast platform);
and carrying out keyword/word recognition and positioning on the speaking data of the user.
It will be appreciated that the above keyword/word recognition may be considered as a search for characters in text, matching, more central to: and (5) positioning.
In this method, positioning is not simply such as: looking at the timestamp of the keyword/word, locating the time it occurred and locating which user made the utterance based on the text-bound user tag class, but:
semantic understanding is further introduced, namely, semantic analysis is carried out on a certain word, a front word and a back word (one word) of the keyword/word, whether the pre-judgment on the player's behavior is triggered is judged, if so, the positioning is successful, and the next step is executed; if not, keyword/word recognition is returned.
Regarding the above semantic parsing, the interpretation is performed with a question-answer function of artificial intelligence, such as: the Chinese is a 'small art' of a mobile phone, when a user wakes up the artificial intelligence small art, the user speaks: how weather today, then its corresponding text/audio results that give weather conditions today; in the process, semantic parsing is included, and the intelligent background needs to understand that the user asks to search weather information.
In one embodiment of the method, a basic semantic parsing scheme: establishing a mapping relation between a large number of words, word combinations and semantic interpretation information, and defining the mapping relation under the semantic interpretation information; when in use, the mapping relation is searched and determined.
When keyword/word recognition and positioning are completed, the following steps are:
based on keyword/word, grasping prejudgement behavior information in user speaking, and based on time sequence logic, comparing score data of subsequent event;
for example: the speaking is that the player A is kicked into the C area to shoot and goal, and the captured prejudgment behavior information is as follows: panel a, panel C, score 1.
Subsequently, counting the correct times of the prejudging actions of the user, and defining the user with the times exceeding a preset first threshold as a reference user;
establishing a user file for a reference user, and recording prejudging behavior information, corresponding event true score data and matched athletes of the prejudging behavior by the user file;
making reference user prejudgement statistics for the present athlete of the current event, and outputting the statistical result as key prompt.
It is understood that the prompting mode may be a popup window, a scrolling text box, or the like of the UI interface.
In one embodiment of the method, the method further comprises:
counting the prejudgment result of the reference user and marking the prejudgment result as a data set A;
comparing the data set A with corresponding event score data, and judging consistency, wherein the result is marked as Km; the data set A adopts a few rules obeying majority, namely, a result with a large duty ratio is identified as a real user prejudging result;
comparing the model prediction result with corresponding event score data, and judging consistency, wherein the result is recorded as Kn;
if the number of the consistent times in the historical Km data set is larger than the number of the consistent times in the Kn data set, defining the user pre-judgment as a characteristic variable, and updating the prediction.
In one embodiment of the method, the method further comprises:
capturing preference trend information (such as abuse, recognition and detraction words) in a reference user utterance based on the keywords/words, and identifying categories of the preference trend information; wherein the categories include forward preferences and negative preferences;
matching preference propensity information with a corresponding athlete based on reference user utterances (chronologic of word occurrences/relationships between words);
when counting data set a, reference users that are negative preferences for the current athlete and whose frequency exceeds a second threshold are skipped.
According to the method, the influence of part of the emotional behaviors of the reference users on the method can be reduced, and the information prompt accuracy is improved.
In one embodiment of the present application, the method further comprises:
and judging the following behavior of the reference user speaking based on the keywords/words, and skipping the reference user when the following behavior is judged to be following the wind, and counting the data set A.
The following wind behavior determination logic may be any of the following:
1) The speaking content is completely consistent with the previous statement;
2) T1 time period (e.g.: 5 s) a large number of (preset third threshold) identical utterances appear in the room, and the content is consistent;
3) Predefined additive vocabulary appears in the speech.
In one embodiment of the present application, if the athlete wears the intelligent sensing device, and the precondition that the physical sign/physiological data of the athlete, especially the time data of the game, are kept secret by each group or organization is not considered, the method further comprises:
establishing each sign/physiological data change curve graph based on the time axis;
extracting physical sign/physiological data change curve graphs of athletes with various score periods, and recording the physical sign/physiological data change curve graphs as score curves; wherein the scoring period is the time elapsed from one score to another score;
determining the change trend (rising Z1 and falling Z2) of the previous n times of score curves of a certain athlete; wherein n, Z1 and Z2 are all arbitrary natural numbers;
if the change trend of the physical sign/physiological data change curve graph of a certain athlete in the current period (from the previous fractional change time point to the current time point) is the same as the change trend of the previous n times of fractional curves, defining the change trend of the current curve as a characteristic variable, and updating the prediction.
The above embodiments are not intended to limit the scope of the present application, so: all equivalent changes in structure, shape and principle of the application should be covered in the scope of protection of the application.
Claims (6)
1. The method for mining and predicting the event key information is characterized by comprising the following steps:
q1, constructing a model foundation, which comprises the following steps:
acquiring multi-field event information as sample data;
analyzing sample data based on pre-recorded event score rules and score factors, and defining characteristic variables required by predictive analysis;
feature extraction is carried out on the analyzed sample data, and a mapping relation between feature variable combination and a scoring result is established based on sequential logic;
defining feature variable combination rules based on the mapping relation and the pre-recorded feature combination examples;
q2, predictive model training, comprising:
collecting and calling past event data of any athlete as a history sample;
extracting features of the historical samples to generate athlete personal specification data;
randomly sampling and dividing the historical samples to obtain a test sample and a training sample;
training and testing a preset prediction model algorithm based on the test sample and the training sample, outputting the preset prediction model algorithm, and storing the preset prediction model algorithm according to the identity of the athlete;
q3, mining event key points, which comprises:
acquiring video data of a current event, and performing face recognition on a top player to obtain identity information of the top player;
according to the identity information, calling a matched prediction model;
acquiring event data of a current event, extracting feature variables, importing a matched prediction model for analysis to obtain a prediction result, and outputting the prediction result serving as a key prompt;
further comprises:
acquiring user speaking data of a pre-docked event online platform;
keyword/word recognition and positioning are carried out on the speaking data of the user;
based on keyword/word, grasping prejudgement behavior information in user speaking, and based on time sequence logic, comparing score data of subsequent event;
counting the correct times of the prejudging actions of the user, and defining the user with the times exceeding a preset first threshold as a reference user;
establishing a user file for a reference user, and recording prejudging behavior information, corresponding event true score data and matched athletes of the prejudging behavior by the user file;
making reference user prejudgment statistics for the present athletes at the current event, and outputting the statistics result to be used as a key prompt;
the positioning includes:
carrying out semantic analysis on a certain keyword/word, speaking front and back, judging whether to trigger the pre-judgment of the player's behavior, if so, positioning successfully, and executing the next step; if not, keyword/word recognition is returned.
2. The event key information mining and prediction method according to claim 1, further comprising:
defining a preamble characteristic variable corresponding to the primary score and combining the characteristic variables into a group of data;
counting each group of data of the same athlete to obtain the same characteristic value and the frequency of occurrence of the characteristic variable combination as the ratio d, and defining the weight value of the characteristic variable based on the frequency of occurrence as the ratio d;
respectively calculating standard deviation/similarity of each characteristic variable;
judging whether the standard deviation/similarity is smaller than a preset first threshold value, if not, calculating the duty ratios of a plurality of different preset value ranges of the same characteristic variable, and recording the duty ratios as a single duty ratio dm;
and synthesizing the weight value and the single-quantity duty ratio dm, obtaining a new weight value and outputting the new weight value.
3. The event key information mining and prediction method according to claim 1, further comprising:
counting the prejudgment result of the reference user and marking the prejudgment result as a data set A;
comparing the data set A with corresponding event score data, and judging consistency, wherein the result is marked as Km;
comparing the model prediction result with corresponding event score data, and judging consistency, wherein the result is recorded as Kn;
if the number of the consistent times in the historical Km data set is larger than the number of the consistent times in the Kn data set, defining the user pre-judgment as a characteristic variable, and updating the prediction.
4. A method of mining and predicting event key information as defined in claim 3, wherein said comparing the data set a with corresponding event score data for consistency determination comprises:
counting the number of various user prejudging results, and comparing the user prejudging result with the largest number with event score data; and if the same, the same is judged as the same.
5. The event key information mining and prediction method according to claim 4, wherein:
capturing preference trend information in a reference user utterance based on keywords/words, and identifying categories of the preference trend information; wherein the categories include forward preferences and negative preferences;
matching preference trend information with a corresponding athlete based on a reference user utterance;
when counting data set a, reference users that are negative preferences for the current athlete and whose frequency exceeds a second threshold are skipped.
6. The event key information mining and prediction method according to claim 1, further comprising: and judging the following behavior of the reference user speaking based on the keywords/words, and skipping the reference user when the following behavior is judged to be following the wind, and counting the data set A.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211520106.5A CN115757551B (en) | 2022-11-30 | 2022-11-30 | Event key information mining and predicting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211520106.5A CN115757551B (en) | 2022-11-30 | 2022-11-30 | Event key information mining and predicting method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115757551A CN115757551A (en) | 2023-03-07 |
| CN115757551B true CN115757551B (en) | 2023-08-25 |
Family
ID=85342799
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211520106.5A Active CN115757551B (en) | 2022-11-30 | 2022-11-30 | Event key information mining and predicting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115757551B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108209902A (en) * | 2017-05-25 | 2018-06-29 | 深圳市未来健身衣科技有限公司 | Sportsman's competitive state appraisal procedure and system |
| CN108905095A (en) * | 2018-08-06 | 2018-11-30 | 百度在线网络技术（北京）有限公司 | Athlete's state evaluating method and equipment |
| CN109165253A (en) * | 2018-08-15 | 2019-01-08 | 宁夏大学 | A kind of method and apparatus of Basketball Tactical auxiliary |
| CN112765949A (en) * | 2021-01-08 | 2021-05-07 | 首都师范大学 | Method, system and storage medium for automatically generating event character live broadcast text |
| CN113571198A (en) * | 2021-08-31 | 2021-10-29 | 平安医疗健康管理股份有限公司 | Conversion rate prediction method, device, equipment and storage medium |
| CN114676314A (en) * | 2021-08-12 | 2022-06-28 | 咪咕音乐有限公司 | Event video recommendation method, device, equipment and storage medium |
| CN114676669A (en) * | 2022-04-15 | 2022-06-28 | 网易（杭州）网络有限公司 | Method and device for generating event abstract text, electronic equipment and storage medium |
-
2022
- 2022-11-30 CN CN202211520106.5A patent/CN115757551B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108209902A (en) * | 2017-05-25 | 2018-06-29 | 深圳市未来健身衣科技有限公司 | Sportsman's competitive state appraisal procedure and system |
| CN108905095A (en) * | 2018-08-06 | 2018-11-30 | 百度在线网络技术（北京）有限公司 | Athlete's state evaluating method and equipment |
| CN109165253A (en) * | 2018-08-15 | 2019-01-08 | 宁夏大学 | A kind of method and apparatus of Basketball Tactical auxiliary |
| CN112765949A (en) * | 2021-01-08 | 2021-05-07 | 首都师范大学 | Method, system and storage medium for automatically generating event character live broadcast text |
| CN114676314A (en) * | 2021-08-12 | 2022-06-28 | 咪咕音乐有限公司 | Event video recommendation method, device, equipment and storage medium |
| CN113571198A (en) * | 2021-08-31 | 2021-10-29 | 平安医疗健康管理股份有限公司 | Conversion rate prediction method, device, equipment and storage medium |
| CN114676669A (en) * | 2022-04-15 | 2022-06-28 | 网易（杭州）网络有限公司 | Method and device for generating event abstract text, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115757551A (en) | 2023-03-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107222865B (en) | Communication swindle real-time detection method and system based on suspicious actions identification | |
| CN109902957B (en) | Data processing method and device | |
| Sadjadi et al. | The 2017 NIST Language Recognition Evaluation. | |
| CN101645088B (en) | Determine the method for auxiliary lexicon, device and the input method system that need to load | |
| CN112533051B (en) | Barrage information display method, barrage information display device, computer equipment and storage medium | |
| CN109640112B (en) | Video processing method, device, equipment and storage medium | |
| CN108920450B (en) | Knowledge point reviewing method based on electronic equipment and electronic equipment | |
| CN108959483B (en) | Search-based learning auxiliary method and electronic equipment | |
| CN108446320A (en) | A kind of data processing method, device and the device for data processing | |
| CN109325115B (en) | Role analysis method and analysis system | |
| CN112765974B (en) | Service assistance method, electronic equipment and readable storage medium | |
| CN109086455B (en) | Method for constructing voice recognition library and learning equipment | |
| Vidaña-Vila et al. | A two-stage approach to automatically detect and classify woodpecker (Fam. Picidae) sounds | |
| CN110765313A (en) | Classified playing method and system for network video barrage | |
| CN109815396A (en) | Search term Weight Determination and device | |
| CN106970906A (en) | A kind of semantic analysis being segmented based on sentence | |
| KR20130022075A (en) | Method for building emotional lexical information and apparatus for the same | |
| CN111841007A (en) | Game control method, device, equipment and storage medium | |
| CN111488501A (en) | E-commerce statistical system based on cloud platform | |
| CN115757551B (en) | Event key information mining and predicting method | |
| CN109766773A (en) | Match monitoring method, device, computer equipment and storage medium | |
| CN108777804B (en) | Media playing method and device | |
| CN114399827B (en) | College graduate career character testing method and system based on facial micro-expression | |
| Brown | Y-ACCDIST: An automatic accent recognition system for forensic applications | |
| CN115147067A (en) | Intelligent recruiter talent recruitment method based on deep learning |
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 |