CN108230007B - User intention identification method and device, electronic equipment and storage medium - Google Patents
User intention identification method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN108230007B CN108230007B CN201711219349.4A CN201711219349A CN108230007B CN 108230007 B CN108230007 B CN 108230007B CN 201711219349 A CN201711219349 A CN 201711219349A CN 108230007 B CN108230007 B CN 108230007B
- Authority
- CN
- China
- Prior art keywords
- user
- feature
- intention
- time
- characteristic
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a method and a device for identifying user intention, electronic equipment and a storage medium, wherein the device comprises the following components: a determination module to determine a plurality of user intents and a plurality of feature items; the extraction module is used for respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention; the first acquisition module is used for acquiring historical behavior data of a user; the training module is used for carrying out model training according to the characteristic numerical values and the historical behavior data of the user so as to obtain the weight value of each characteristic item; the second acquisition module is used for acquiring a plurality of current characteristic items, and the plurality of current characteristic items respectively have real-time characteristic numerical values; the intention probability distribution calculation module is used for weighting and summing the real-time characteristic numerical values according to the weight values so as to obtain the current probability distribution of the plurality of user intentions, and the problem that the user instantaneous tendency cannot be reflected under the condition of no query in the prior art is solved, so that the user intentions have comparability.
Description
Technical Field
The present invention relates to the field of internet technologies, and in particular, to a method and an apparatus for identifying a user intention, an electronic device, and a storage medium.
Background
With the increasing maturity of mobile internet technology, people can conveniently access the internet through mobile equipment at any time and place, the demand of users on information is also increasingly personalized, and the information demand of users at the moment, namely the real-time user intention, is an important aspect for realizing personalization.
In the existing comprehensive search recommendation system under the O2O (Online To Offline ) scene, the intention of a user is generally added To a learning sequencing model by using historical behaviors such as clicking and ordering as features, and services To be recommended To the user are obtained through training of the model, but the recommendation method can only reflect the historical preference of the user for a certain class of services. In addition, the search system mostly depends on the input query of the user besides using the historical behavior of the user, and obtains a corresponding recommendation result by matching the text input by the user. However, the instantaneous tendency of the user, e.g., the user's intention or tendency at the time of initial start-up of the search system, is not well reflected without the query. Meanwhile, a plurality of systems need to access the historical behavior characteristics of the user for a plurality of times, so that the complexity of model calculation is increased, and the error probability of the systems is increased.
Disclosure of Invention
In view of the above problems, embodiments of the present invention are provided to provide a user intention identification method, a user intention identification apparatus, an electronic device, and a storage medium corresponding thereto that overcome or at least partially solve the above problems.
In order to solve the above problem, an embodiment of the present invention discloses an apparatus for identifying a user intention, including:
a determination module to determine a plurality of user intents and a plurality of feature items;
the extraction module is used for respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention;
the first acquisition module is used for acquiring historical behavior data of a user;
the training module is used for carrying out model training according to the characteristic numerical values and the historical behavior data of the user so as to obtain the weight values of all the characteristic items;
the second acquisition module is used for acquiring a plurality of current characteristic items, and the plurality of current characteristic items respectively have real-time characteristic numerical values;
and the intention probability distribution calculation module is used for weighting and summing the real-time characteristic numerical values according to the weight values so as to obtain the current probability distribution of the user intentions.
Optionally, the plurality of feature items include a temporal feature item, the temporal feature item corresponding to a feature value of each user intention being converted into a temporal feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items include category feature items, the category feature items corresponding to feature values of each user intention are converted into a category feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items includes a time-cumulative feature item, and the time-cumulative feature item is converted into a time-cumulative feature item hash function corresponding to a feature value of each user intention.
Optionally, when the plurality of feature items include a temporal feature item, the extracting module includes:
a dividing submodule for dividing the unit time period into a plurality of time slices;
the first statistic submodule is used for respectively counting the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices;
and the first generation submodule is used for generating the time characteristic item hash function according to the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices.
Optionally, when the plurality of feature items include category feature items, the extracting module includes:
the second counting submodule is used for respectively counting the frequency of the intention of each user in any category;
and the second generation submodule is used for carrying out normalization and smoothing processing on the frequency of each user intention under any category to generate the category characteristic item hash function.
Optionally, when the plurality of feature items includes an age-cumulative feature item, the extracting module includes:
the obtaining submodule is used for obtaining the occurrence time of each user intention of the current user, which is the latest time from the current time, and generating timeliness parameters;
the third counting submodule is used for respectively counting the maximum value of the times of occurrence of each user intention in a preset time period and the times of occurrence of the user intention of the current user in the preset time period;
a third generation submodule, configured to generate an accumulative parameter according to a maximum value of the number of times that each user intention occurs within a preset time period and the number of times that a current user intends to occur within the preset time period;
and the fourth generation submodule is used for generating the time effect-cumulative characteristic item hash function by adopting the time effect parameter and the cumulative parameter.
Optionally, the timeliness parameter is generated by taking a logarithm of the occurrence time of each user intention of the current user that is the closest to the current time and translating; and/or the presence of a gas in the gas,
the cumulative parameter is generated by taking a hyperbolic tangent as a ratio of a number of times that the current user has the user intention within the preset time period to a maximum of a number of times that each of the user intentions has occurred within the preset time period.
Optionally, the training module comprises:
and the training submodule is used for performing model training on the time characteristic item hash function, the category characteristic item hash function and/or the aging-accumulation characteristic item hash function according to the historical behavior data of the user so as to obtain the weight value of each characteristic item.
Optionally, the method further comprises:
and the identification module is used for identifying the user intention corresponding to the maximum probability value in the probability distribution as the target user intention.
In order to solve the above problem, an embodiment of the present invention discloses a method for identifying a user intention, including:
determining a plurality of user intents and a plurality of feature items;
respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention;
acquiring historical behavior data of a user;
performing model training according to the characteristic numerical values and the historical behavior data of the user to obtain the weight values of the characteristic items;
acquiring a plurality of current characteristic items, wherein the plurality of current characteristic items respectively have real-time characteristic numerical values;
and carrying out weighted summation on the real-time characteristic numerical values according to the weight values so as to obtain the current probability distribution of the user intentions.
Optionally, the plurality of feature items include a temporal feature item, the temporal feature item corresponding to a feature value of each user intention being converted into a temporal feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items include category feature items, the category feature items corresponding to feature values of each user intention are converted into a category feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items includes a time-cumulative feature item, and the time-cumulative feature item is converted into a time-cumulative feature item hash function corresponding to a feature value of each user intention.
Optionally, when the plurality of feature items include a temporal feature item, the step of respectively extracting a feature value corresponding to each user intention for each feature item includes:
dividing a unit time period into a plurality of time slices;
respectively counting the maximum value of the number of times of each user intention in each time slice and the number of times of each user intention in all time slices;
and generating the time characteristic item hash function according to the maximum value of the number of times of occurrence of each user intention in each time slice and the number of times of occurrence of each user intention in all time slices.
Optionally, when the plurality of feature items include category feature items, the step of respectively extracting a feature value corresponding to each user intention for each feature item includes:
respectively counting the number of times of the intention of each user in any category;
and carrying out normalization and smoothing treatment on the times of each user intention under any category to generate the category characteristic item hash function.
Optionally, when the plurality of feature items includes an age-cumulative feature item, the step of separately extracting a feature value corresponding to each user intention for each feature item includes:
acquiring the occurrence time of each user intention of the current user, which is the latest time from the current time, and generating timeliness parameters;
respectively counting the maximum number of times of occurrence of each user intention in a preset time period, and the number of times of occurrence of the user intention of the current user in the preset time period;
generating accumulative parameters according to the maximum value of the times of occurrence of each user intention in a preset time period and the times of occurrence of the user intention of the current user in the preset time period;
and generating the time efficiency-cumulative characteristic item hash function by adopting the time efficiency parameter and the cumulative parameter.
Optionally, the timeliness parameter is generated by taking a logarithm of the occurrence time of each user intention of the current user that is the closest to the current time and translating; and/or the presence of a gas in the gas,
the cumulative parameter is generated by taking a hyperbolic tangent as a ratio of a number of times that the current user has the user intention within the preset time period to a maximum of a number of times that each of the user intentions has occurred within the preset time period.
Optionally, the step of performing model training according to the feature value and the user historical behavior data to obtain a weight value of each feature item includes:
and performing model training on a time characteristic item hash function, a category characteristic item hash function and/or an aging-accumulation characteristic item hash function according to the historical behavior data of the user to obtain the weight value of each characteristic item.
Optionally, after the step of obtaining the current probability distribution of the plurality of user intentions, further comprising:
and identifying the user intention corresponding to the maximum probability value in the probability distribution as a target user intention.
In order to solve the above problem, an embodiment of the present invention discloses an electronic device, including:
a processor; and the number of the first and second groups,
a memory for storing executable instructions of the processor;
wherein the processor is configured to perform the steps of the above-mentioned user intention identification method by executing the executable instructions.
In order to solve the above problem, an embodiment of the present invention discloses a storage medium on which a computer program is stored, which when executed by a processor, implements the steps of the above-described recognition method of the user's intention.
Compared with the background art, the embodiment of the invention has the following advantages:
according to the embodiment of the invention, a plurality of user intentions and a plurality of feature items are determined, the feature value of each feature item corresponding to each user intention is extracted respectively, model training is carried out according to the feature value and the user historical behavior data on the basis of obtaining the user historical behavior data, and the weight value of each feature item is obtained, so that after the current plurality of feature items are obtained, the real-time feature values of the current plurality of feature items can be weighted and summed according to the weight values, and the current probability distribution of the plurality of intentions of the user is obtained. The embodiment of the invention can better reflect the intention tendency of the user at the request moment by calculating the probability distribution of the user intention, and ensure that each user intention has comparability and can be transversely compared.
Drawings
FIG. 1 is a flow chart of the steps of a method for identifying user intent, in accordance with one embodiment of the present invention;
FIG. 2 is a flow chart of steps of another method of identifying user intent, in accordance with one embodiment of the present invention;
fig. 3 is a block diagram of a user intention recognition apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Referring to fig. 1, a flowchart illustrating steps of a method for identifying a user intention according to an embodiment of the present invention is shown, and specifically may include the following steps:
it should be noted that the embodiment of the present invention may be applied to a client, where the client may be connected to a server or a server cluster of a third party, such as a distributed system, and may capture service data of a service object in a network platform, and the network platform may be an independent server or a server cluster and is used to perform service processing on the service object.
Different business objects can be provided in different business fields, for example, in the communication field, the business object can be communication data; in the news media field, the business object may be news data; in the search field, the business object may be a web page; in the field of electronic commerce, business objects may be goods or types of goods, and so on.
In the embodiment of the invention, the user intention can be the intention of the user aiming at each business object. For example, if the business objects include restaurant to store and take out, the user intent may be an intent that the user tends to select restaurant to store or take out, respectively.
In an embodiment of the present invention, the plurality of feature items may include a client condition feature or a user behavior feature. Such as time, weather, user clicks, browsing, or ordering activities, etc. The present embodiment does not limit the business object and feature item targeted by the user's intention.
102, respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention;
in the embodiment of the present invention, the feature value may refer to a probability value of each user intention under each feature item. For example, for a user intent of take-away and a feature item of time, a probability value for each hour of take-away within 24 hours of the day may be extracted.
In particular implementations, the plurality of feature items may be classified according to respective features. For example, for the feature items that are only closely related to time, the feature items can be divided into time feature items; for the feature items which are only closely related to the category, the feature items can be divided into category feature items; for those feature items that need to consider not only the aging property but also the accumulation property, they can be classified as aging-accumulation feature items.
In the embodiment of the present invention, the feature value of each time feature item corresponding to each user intention may be converted into a time feature item hash function; the feature value of each category feature item corresponding to each user intention can be converted into a category feature item hash function; and the feature value of each aging-cumulative feature item corresponding to each user intention may be converted into an aging-cumulative feature item hash function.
103, acquiring historical behavior data of a user;
in the embodiment of the present invention, the user historical behavior data may refer to past operation behavior data of the user. For example, the user may click, browse, or place orders for business objects that the user intends to target.
104, performing model training according to the characteristic numerical values and the historical behavior data of the user to obtain the weight values of the characteristic items;
in a specific implementation, after determining a feature value of each feature item corresponding to each user intention, model training may be performed based on user historical behavior data to obtain a weight value of each feature item.
It should be noted that the calculated weight value of each feature item is not changed when the subsequent user intention prediction is performed.
105, acquiring a plurality of current characteristic items, wherein the plurality of current characteristic items respectively have real-time characteristic numerical values;
the current plurality of feature items may refer to all features currently possessed. Such as the current time, location, weather, etc. The plurality of characteristic items at present may each have a real-time characteristic value.
And 106, carrying out weighted summation on the real-time characteristic numerical values according to the weight values so as to obtain the current probability distribution of the user intentions.
In the embodiment of the present invention, when identifying the user intention, the real-time feature data of the current plurality of feature items may be weighted and summed according to the weight value of each feature item, and the result of the weighted summation is the current probability distribution of each user intention.
The current probability distribution of the respective user intentions may represent the magnitude of the user's tendency for each user intention. For example, the greater the probability value, the greater the tendency of the user to indicate the intent.
In the embodiment of the invention, a plurality of user intentions and a plurality of feature items are determined, a feature value corresponding to each user intention of each feature item is extracted respectively, model training is carried out according to the feature value and the user historical behavior data on the basis of obtaining user historical behavior data, and a weight value of each feature item is obtained, so that after a plurality of current feature items are obtained, the real-time feature values of the plurality of current feature items can be weighted and summed according to the weight values, and the current probability distribution of the plurality of intentions of the user is obtained. The embodiment of the invention can better reflect the intention tendency of the user at the request moment by calculating the probability distribution of the user intention, and ensure that each user intention has comparability and can be transversely compared.
Referring to fig. 2, a flowchart illustrating steps of another method for identifying a user intention according to an embodiment of the present invention is shown, which may specifically include the following steps:
in the embodiment of the invention, the user intention may refer to the intention or tendency of the user for each business object, and the corresponding business object may be recommended to the user by determining the intention of the user.
In the embodiment of the present invention, in order to recommend a service object to a user, a plurality of service objects to be recommended and feature items for analyzing the plurality of service objects may be determined first, so as to determine a probability that the user opens a certain application APP to a certain service object at the moment.
In particular implementations, the plurality of user intentions may be a set of intentions { P } consisting of a plurality of intentions of the user1，P2，P3，…PNSuch as { catering to stores, take-out, movies, hotels, travel, KTV, entertainment } and the like. Meanwhile, a binary variable y is set to be e {0, 1}, and if the user has a certain intention, the value of y is 1; if the user does not have some intent, then the value of y is 0. The intensity of a user's intention can thus be expressed as the probability that y is 1: p { y ═ 1 }.
The strength of this intention can be expressed as P { y ═ 1| open }, at the moment when the user opens some APP, where open represents the event that the user opens the APP.
In the embodiment of the present invention, the plurality of feature items may include objective condition features or user behavior features. The objective condition characteristics may refer to objective conditions such as environmental factors and time factors. Such as time, weather, location, etc. The user behavior characteristics may refer to subjective behaviors of the user such as browsing, clicking, placing orders and the like for each business object.
In a specific implementation, a plurality of feature items may also exist in the form of a feature item set. For example, { time, weather, location, personal history ordering, personal history queries } and the like.
in the embodiment of the present invention, after introducing the objective condition characteristic and the user behavior characteristic, the strength P { y ═ 1| open } of the user for a certain intention may be further expressed as:
P{y＝1|open}＝∑C∈{Context，User}P{y＝1|C，open}P{C|open}……(1)
where C ∈ { Context, User } can be understood as C being one of the set of objective condition features and User behavior features.
Therefore, calculating the probability of the user opening the APP at the moment can be converted into calculating the sum of the probabilities that the user has the APP at the moment when the user opens the APP under all the characteristic items.
In a specific implementation, an independence assumption may be introduced: let the instant a user opens an APP have a certain feature independent of the probability under that feature that the user has a certain intention, and can thus be calculated in the form of a probability product. In fact, it is not necessarily independent of some intention that the user has some feature at the moment of opening the APP, but this independence assumption is statistically reasonable and can simplify the model calculations.
In the embodiment of the present invention, assuming that the probability that the user has a certain feature when opening the APP is θ, and the probability that the user has a certain intention under the feature is x, the above equation (1) may be converted into a unitary logistic regression problem:
P{y＝1|open}＝sigmoid(∑iθi*xi)，xi∈[0，1]……(2)
therefore, calculating the distribution of user intentions is calculating all { P }1，P2，P3，…，PNIn which P isi∈[0，1]，∑Pi＝1。
In conjunction with the above derivation, the problem of solving the probability distribution of user intent can be transformed to solve the Softmax regression problem, whose formula is shown below:
where j e {1, 2, …, N } represents N intents, xiIs a feature vector which comprises objective condition features and user behavior features.
For the formula (4), θ can be obtained by learning the sample through a machine learning algorithm, so that when the instantaneous intention of a user to open the APP needs to be predicted, only a plurality of current features need to be obtained and then the current features are brought into the formula (4) with θ obtained, and the probability distribution of the user for each user intention can be calculated.
In the embodiment of the present invention, before the solution is performed, a feature value corresponding to each user intention of each feature item may be extracted first.
In particular implementations, the plurality of feature items may be classified according to respective features. Such as temporal feature terms, category feature terms, and age-cumulative feature terms. The time characteristic item may include time; the category characteristic items can comprise characteristics of weather, places and the like; and the aging-cumulative feature item can comprise features such as personal history ordering, personal history query and the like.
In the embodiment of the present invention, the feature value of each time feature item corresponding to each user intention may be converted into a time feature item hash function; the feature value of each category feature item corresponding to each user intention can be converted into a category feature item hash function; and the feature value of each aging-cumulative feature item corresponding to each user intention may be converted into an aging-cumulative feature item hash function.
In a specific implementation, for the time feature item, the unit time period may be first divided into a plurality of time slices, each time slice may be denoted as TiThen, each time slice T is counted separatelyiCount (e, T) of the number of times each user intends to occuri) And a maximum value max of the number of times each user intends to occur in all time slices (e, T)i) }; so that each time slice T can be based oniCount (e, T) of the number of times each user intends to occuri) And a maximum value max of the number of times each user intends to occur in all time slices (e, T)i) And generating a time characteristic item hash function, wherein the formula is as follows:
specifically, the number of times a certain user's intention occurred at different periods in the day may be examined in a unit time period of 24 hours a day. For example, every 5 minutes is taken as a time interval, 24 hours in a day is set as 288 time slices, each time slice is 5 minutes, and then the obtained hash function values are averaged by counting the values in the last three months, and the average value is the final value.
In practical application, when the user opens the APP at any time of a day, the characteristic value of the user intention corresponding to the time can be obtained through the set of hash functions.
For example, for the user's intention of taking a bill out, it can be known through the hash function that 11-13 points per day are the times of day when the feature is the strongest.
For category feature items such as weather and places, the corresponding feature values can be obtained by counting the occurrence times of the intentions of the users in the categories.
In a specific implementation, the number of times count (e, C) of each user intention occurrence under any category can be counted respectivelyi) Then count (e, C) the number of times each user's intention occurred under any of the above categoriesi) And performing normalization and smoothing processing to generate a category feature item hash function, wherein the smoothing processing can be performed by adopting a tanh function. The corresponding formula can be expressed as follows:
by adopting the formula, a group of hash functions corresponding to the category feature items can be obtained, so that the intensity of the user intention of the user in a certain category scene when the APP is opened by the user can be obtained, namely the intensity of the intention of the user to a certain service object in the category.
Furthermore, for some types of feature items, not only the timeliness but also the cumulation need to be considered. For example, for the behavior of buying takeaway by the user, if the user a has a record of taking out the order within one week and the user B has a record of taking out the order three weeks ago, the strength of the takeaway intention of the user a and the user B may be different, which is a time-efficient embodiment. In addition, if user a places a take 30 and user B only places 5 in a period, such as three months, the intentions of users a and B for take should be different, which is a cumulative embodiment. The characteristic items such as personal historical ordering and personal historical inquiry are the characteristic items which need to consider timeliness and accumulation at the same time.
In the embodiment of the present invention, the aging performance can be recorded as x for the aging-accumulation characteristic term1Cumulative score x2The strength of the corresponding feature may then be expressed as:
y＝α*x1+(1-α)*x2，α∈[0，1]……(8)
wherein α is a weighting parameter of timeliness and accumulativity, the value range is [0, 1], and the magnitude of α can be specifically determined according to actual needs, which is not limited in the embodiment of the present invention.
In the embodiment of the invention, the occurrence time of each user intention of the current user, which is the latest time from the current time, can be obtained to generate the timeliness parameter.
In a specific implementation, the timeliness parameter may be generated by taking a logarithm of the occurrence time of each user intention of the current user that is the closest to the current time and translating the logarithm.
For example, a time decay function may be used to record a value corresponding to the most recent time a user intended to have occurred as x1The value of (c). Wherein the time decay function can be obtained by translating the origin by using a log function.
Then, the maximum value of the number of times that each user intention occurs within a preset time period, and the number of times that the user intention occurs within the preset time period of the current user can be respectively counted. Accordingly, the accumulative parameter can be generated according to the maximum value of the times of occurrence of each user intention in the preset time period and the times of occurrence of the user intention of the current user in the preset time period.
In a specific implementation, the accumulative parameter may be generated by taking a hyperbolic tangent (tanh) as a ratio of the number of times that the current user has a user intention within a preset time period to a maximum of the number of times that each user intention has occurred within the preset time period.
For example, the maximum value of the number of times that all users have the respective user intentions within a preset time period may be counted, and the maximum value is recorded as max { count (u, e, T) }, and a function is established with this as the segmentation point:
where u (e) is the number of times a current user has a certain user intention within the same time period.
Then, using tanh function to normalize u (e) and max, obtaining x2The value of (c):
after the timeliness parameter and the accumulative parameter are obtained respectively, the timeliness parameter and the accumulative parameter can be adopted to generate the timeliness-accumulative feature item hash function.
in the embodiment of the present invention, the user historical behavior data may refer to past operation behavior data of the user. For example, the user may click, browse, or place orders for business objects that the user intends to target.
In the embodiment of the invention, the training sample set can be generated according to the extracted historical behavior data of the user. The training sample set may include a positive sample set and a negative sample set.
In a specific implementation, after a user opens an APP, for a plurality of business objects displayed by the APP, if the user clicks a certain business object, a class corresponding to the business object may be used as a positive sample set, and classes corresponding to other business objects that the user does not click may be used as a negative sample set.
Or, if the search keyword displayed by the APP default is directly searched by the user, the recommendation of the search keyword can be considered to be accurate, so that the classification corresponding to the keyword can be used as a positive sample set. Conversely, if the user does not search for the default presented keyword, it may be used as a negative sample set. Of course, those skilled in the art may specifically adjust the data source of the sample set according to actual needs, which is not limited in this embodiment.
in the embodiment of the present invention, after the training sample set is obtained, model training may be performed on the time feature item hash function, the category feature item hash function, and the age-cumulative feature item hash function obtained in step 202, so as to obtain a weight value of each feature item.
In a specific implementation, the feature value of the extracted feature item may be substituted into formula (4), and the weight value of the corresponding feature item, that is, the value of θ in formula (4), may be obtained.
in the embodiment of the present invention, the probability of the user for each user intention may be a result of the multiple feature items acting together with the corresponding weight values, and therefore, when determining the probability distribution of the multiple user intentions, the current multiple feature items may be obtained first.
The current plurality of feature items may refer to all features currently possessed. Such as the current time, location, weather, etc. The plurality of characteristic items at present may each have a real-time characteristic value.
In a specific implementation, the real-time feature values corresponding to the current multiple feature items may be determined according to step 202, which is not described in detail in this embodiment.
in the embodiment of the present invention, after obtaining a plurality of current feature items and determining a real-time feature value corresponding to each feature item, the weighting values calculated in step 204 may be adopted to perform weighted summation on the real-time feature values of the plurality of current feature items, so as to obtain the current probability distribution of the plurality of user intentions.
In a specific implementation, the current real-time feature values of a plurality of feature items can be substituted into the formula (4) to be solved, and the current probability distribution of a plurality of user intentions can be directly output.
And step 207, identifying the user intention corresponding to the maximum probability value in the probability distribution as the target user intention.
In the embodiment of the invention, the probability distribution of the user aiming at each user intention reflects the strength of the intention of the user to each business object. Therefore, after obtaining the probability distribution of the plurality of user intentions, the target user intention may be further extracted from the plurality of user intentions.
For example, the user intention corresponding to the maximum probability value in the probability distribution may be extracted as the target user intention, and the service object corresponding to the target user intention is the target service object with the strongest user intention.
In the embodiment of the invention, after the intention of the target user is determined, the target business object corresponding to the intention of the target user can be recommended to the user.
Specifically, the target business object can be displayed in the search box, so that the user can conveniently and directly search by adopting the target business object; and the target business object can be preferentially displayed in a search result page of the user so as to realize personalized sequencing of the search result. Of course, those skilled in the art may select other recommendation manners according to actual needs, and this embodiment is not limited to this.
In the embodiment of the invention, the problem that the instant tendency of the user cannot be reflected under the condition of no query in the prior art is solved by calculating the probability distribution of the user to the user intentions and identifying the target user intention from the user intentions, the intention tendency of the user at the request moment is better reflected by calculating the probability distribution of the user intentions, and each user intention has comparability and can be compared transversely. Secondly, when the embodiment of the invention recommends the service object, the intention of the user can be predicted at a certain moment without depending on the query or other operations of the user, and the effectiveness of recommending the service object is improved.
For the convenience of understanding, the following describes the method for identifying the user intention of the present invention with a complete example.
S1, in the O2O scene, the user can use the application program APP for consumption. In order to quickly predict the user's intention when the user opens the APP, so as to better recommend the consumption type (i.e. business object) meeting the user's intention to the user, an intention set and a feature item set may be first determined, for example, the intention set may be P e { restaurant to store, take out, movie, hotel, travel, KTV, entertainment }, and the feature item set may be C e { time, weather, place, personal history ordering, personal history query }, wherein time, weather, place belong to objective condition features, personal history ordering and personal history query belong to user behavior features, and total five-dimensional features.
And S2, selecting a proper feature model according to the determined feature items to extract feature values.
For example, for the time dimension, a time feature item model can be adopted to measure ordering events for all users classified according to each intention, and a numerical feature hash function of a user group at different time intervals in one day is calculated, so that the intentions of the catering class have higher weight in the noon, the strong intention of the movie appears in the gold time in the evening, and the strong intention of the hotel is 22 o' clock behind the evening.
For weather and location feature items, a category feature item model is suitably used. Further, since severe weather conditions will affect consumption behaviors related to life services, the weather can be classified into { low temperature, heavy rain, medium rain, light rain, strong wind, heavy haze, heavy fog, thunder, snow, and others }, and the weather factors in the normal range are placed into other categories, which are considered to have little effect on the consumption behaviors of the user; meanwhile, one day can be divided into { morning, afternoon and evening }, and one week can be divided into { working day and resting day }. Then, the classifications are combined respectively, and a category feature item model is adopted to calculate a hash function, and the set of hash functions can be used for off-line training models and weather feature calculation during on-line prediction. Similar to the weather feature item, the location feature item can be further classified into { working place, residence place, business, route, and other }, a week is classified into { working day, holiday }, and then the classifications are combined to obtain a hash function of the location feature item.
For the characteristic items of the individual historical order and the individual historical query, an aging-cumulative characteristic item model can be adopted to comprehensively calculate the numerical scores of the characteristic items from the aspects of timeliness and cumulation. For personal history ordering, the ordering condition of each user under each intention classification in the last three months can be measured; and the timeliness can be measured using the function (4+ log)0.5(t +1))/4, whose numerical characteristic is that y approaches 1 as t approaches 0 on the horizontal axis and progressively approaches the horizontal axis as t increases, i.e. the value 0, the function characterizes the closer the time to order, the more interest tends to a certain intention category and the further the time, the less interest decays. In specific calculation, the timeliness numerical value can be updated by adopting the latest order, and the cumulativity can be implemented by using the maximum order value of the user in three months as a segmentation function, so that normalized numerical scores are made for the orders of all intention categories in three months of each user. Thus, it can be seen that the more orders are placed, the higher the score. For personal historical query feature items, similar to the ordering described above, data calculations within 4 weeks may be selected. The feature term dependent query classifier classifies each search query, and statistics of the number of queries of each intention category of the user can be obtained through calculation, so that a score is obtained. It can be seen that the more the number of queries, the closer the query time, the higher the score.
In the feature item set, objective condition feature items such as time, place, weather and the like are considered to be behaviors of user groups, and user behavior feature items such as personal history ordering and personal history inquiry are considered to be behaviors of user individuals. After the above feature values are obtained, model training may be started.
S3, specifically, a Softmax regression model may be used for training, so that the value of θ in equation (4) may be obtained. In a specific implementation, for the intention set of P e { restaurant to store, take out, movie, hotel, tour, KTV, entertainment }, and the feature item set of C e { time, weather, place, personal history order, personal history query }, the example can be represented as a 7 x 5 matrix during model training, so that the theta parameter can be obtained by using the L-BFGS algorithm.
S4, when the user opens the APP, the feature information corresponding to each feature item at present, such as the current time, weather and the like, can be obtained, so as to continue to calculate the feature value corresponding to the current feature item according to the above-described feature extraction method, and input the trained Softmax regression model, so as to obtain the probability distribution of the user to the intention set, wherein the probability distribution reflects the strength of the intention of the user to each business object.
S5, after obtaining the real-time intention distribution of the user, selecting the recommendation query in the category with the strongest intention in the application recommended by the default search word, and recommending the service object corresponding to the maximum probability value to the user.
It should be noted that, for simplicity of description, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the illustrated order of acts, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are presently preferred and that no particular act is required to implement the invention.
Referring to fig. 3, a block diagram of a structure of an embodiment of a service object recommendation apparatus of the present invention is shown, which may specifically include the following modules:
a determining module 301 for determining a plurality of user intents and a plurality of feature items;
an extraction module 302, configured to extract a feature value corresponding to each user intention for each feature item respectively;
a first obtaining module 303, configured to obtain user historical behavior data;
a training module 304, configured to perform model training according to the feature value and the user historical behavior data to obtain a weight value of each feature item;
a second obtaining module 305, configured to obtain a plurality of current feature items, where the plurality of current feature items respectively have real-time feature values;
an intention probability distribution calculating module 306, configured to perform weighted summation on the real-time feature values according to the weight values, so as to obtain a current probability distribution of the user intentions.
In an embodiment of the present invention, the plurality of feature items include a temporal feature item, and the feature value of the temporal feature item corresponding to each user intention is converted into a temporal feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items include category feature items, the category feature items corresponding to feature values of each user intention are converted into a category feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items includes a time-cumulative feature item, and the time-cumulative feature item is converted into a time-cumulative feature item hash function corresponding to a feature value of each user intention.
In this embodiment of the present invention, when the feature items include a time feature item, the extracting module 302 may specifically include the following sub-modules:
a dividing submodule for dividing the unit time period into a plurality of time slices;
the first statistic submodule is used for respectively counting the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices;
and the first generation submodule is used for generating the time characteristic item hash function according to the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices.
In this embodiment of the present invention, when the feature items include category feature items, the extracting module 302 may specifically include the following sub-modules:
the second counting submodule is used for respectively counting the frequency of the intention of each user in any category;
and the second generation submodule is used for carrying out normalization and smoothing processing on the frequency of each user intention under any category to generate the category characteristic item hash function.
In this embodiment of the present invention, when the plurality of feature items include an age-cumulative feature item, the extracting module 302 may specifically include the following sub-modules:
the obtaining submodule is used for obtaining the occurrence time of each user intention of the current user, which is the latest time from the current time, and generating timeliness parameters;
the third counting submodule is used for respectively counting the maximum value of the times of occurrence of each user intention in a preset time period and the times of occurrence of the user intention of the current user in the preset time period;
a third generation submodule, configured to generate an accumulative parameter according to a maximum value of the number of times that each user intention occurs within a preset time period and the number of times that a current user intends to occur within the preset time period;
and the fourth generation submodule is used for generating the time effect-cumulative characteristic item hash function by adopting the time effect parameter and the cumulative parameter.
In the embodiment of the invention, the timeliness parameter is generated by taking the logarithm of the occurrence time of each user intention of the current user, which is closest to the current time, and translating the logarithm; and/or the presence of a gas in the gas,
the cumulative parameter is generated by taking a hyperbolic tangent as a ratio of a number of times that the current user has the user intention within the preset time period to a maximum of a number of times that each of the user intentions has occurred within the preset time period.
In this embodiment of the present invention, the training module 304 may specifically include the following sub-modules:
and the training submodule is used for performing model training on the time characteristic item hash function, the category characteristic item hash function and/or the aging-accumulation characteristic item hash function according to the historical behavior data of the user so as to obtain the weight value of each characteristic item.
In the embodiment of the present invention, the apparatus may further include the following modules:
and the identification module is used for identifying the user intention corresponding to the maximum probability value in the probability distribution as the target user intention.
For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
The embodiment of the invention also provides an electron, which comprises a processor; and the number of the first and second groups,
a memory for storing executable instructions of the processor;
the processor is configured to execute each process of the embodiment of the method for identifying a user intention by executing the executable instruction, and can achieve the same technical effect, and the details are not repeated here to avoid repetition.
The embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the computer program implements each process of the above-mentioned embodiment of the method for identifying a user intention, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or terminal that comprises the element.
The above detailed description is provided for a method for identifying a user intention, an apparatus for identifying a user intention, an electronic device and a storage medium, and specific examples are applied in this text to explain the principles and embodiments of the present invention, and the descriptions of the above embodiments are only used to help understanding the method and the core ideas of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (14)
1. An apparatus for recognizing a user's intention, comprising:
a determination module to determine a plurality of user intents and a plurality of feature items;
the extraction module is used for respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention;
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring historical behavior data of a user, and the historical behavior data of the user is historical behavior data of the user for a business object which each user intends to aim at;
the training module is used for carrying out model training according to the characteristic numerical values and the historical behavior data of the user so as to obtain the weight value of each characteristic item;
the second acquisition module is used for acquiring a plurality of current characteristic items, and the plurality of current characteristic items respectively have real-time characteristic numerical values;
and the intention probability distribution calculation module is used for weighting and summing the real-time characteristic numerical values according to the weight values so as to obtain the current probability distribution of the intentions of the plurality of users on the premise of not depending on the query or other operations of the users, so as to realize the real-time prediction of the intention distribution of the users aiming at different business objects, wherein the probability distribution is used for reflecting the instantaneous tendency of the users under the condition of no query.
2. The apparatus of claim 1,
the plurality of feature items include a temporal feature item, the temporal feature item being converted into a temporal feature item hash function corresponding to a feature value of each user intention; and/or the presence of a gas in the gas,
the plurality of feature items include category feature items, the category feature items corresponding to feature values of each user intention are converted into a category feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items includes a time-cumulative feature item, and the time-cumulative feature item is converted into a time-cumulative feature item hash function corresponding to a feature value of each user intention.
3. The apparatus of claim 2, wherein when the plurality of feature items includes a temporal feature item, the extraction module comprises:
a dividing submodule for dividing the unit time period into a plurality of time slices;
the first statistic submodule is used for respectively counting the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices;
and the first generation submodule is used for generating the time characteristic item hash function according to the maximum value of the frequency of each user intention in each time slice and the frequency of each user intention in all time slices.
4. The apparatus of claim 2, wherein when the plurality of feature items comprise category feature items, the extraction module comprises:
the second counting submodule is used for respectively counting the frequency of the intention of each user in any category;
and the second generation submodule is used for carrying out normalization and smoothing processing on the frequency of each user intention under any category to generate the category characteristic item hash function.
5. The apparatus of claim 2, wherein when the plurality of feature terms comprises an age-cumulative feature term, the extraction module comprises:
the obtaining submodule is used for obtaining the occurrence time of each user intention of the current user, which is the latest time from the current time, and generating timeliness parameters;
the third counting submodule is used for respectively counting the maximum value of the times of occurrence of each user intention in a preset time period and the times of occurrence of the user intention of the current user in the preset time period;
a third generation submodule, configured to generate an accumulative parameter according to a maximum value of the number of times that each user intention occurs within a preset time period and the number of times that a current user intends to occur within the preset time period;
and the fourth generation submodule is used for generating the time effect-cumulative characteristic item hash function by adopting the time effect parameter and the cumulative parameter.
6. The apparatus of claim 5,
the timeliness parameter is generated by taking the logarithm of the occurrence time of each user intention of the current user, which is closest to the current time, and translating the logarithm; and/or the presence of a gas in the gas,
the cumulative parameter is generated by taking a hyperbolic tangent as a ratio of a number of times that the current user has the user intention within the preset time period to a maximum of a number of times that each of the user intentions has occurred within the preset time period.
7. The apparatus of claim 2, wherein the training module comprises:
and the training submodule is used for performing model training on the time characteristic item hash function, the category characteristic item hash function and/or the aging-accumulation characteristic item hash function according to the historical behavior data of the user so as to obtain the weight value of each characteristic item.
8. The apparatus of claim 1, further comprising:
and the identification module is used for identifying the user intention corresponding to the maximum probability value in the probability distribution as the target user intention.
9. A method for recognizing user intention is characterized by comprising the following steps:
determining a plurality of user intents and a plurality of feature items;
respectively extracting a characteristic numerical value of each characteristic item corresponding to each user intention;
acquiring historical behavior data of a user, wherein the historical behavior data of the user is the historical behavior data of the user on the service object which is intended by each user;
performing model training according to the characteristic numerical values and the historical behavior data of the user to obtain weight values of all characteristic items;
acquiring a plurality of current characteristic items, wherein the plurality of current characteristic items respectively have real-time characteristic numerical values;
and according to the weight values, carrying out weighted summation on the real-time characteristic numerical values so as to obtain the current probability distribution of the intentions of the plurality of users on the premise of not depending on the query or other operations of the users, so as to realize the real-time prediction of the intention distribution of the users aiming at different business objects, wherein the probability distribution is used for reflecting the instantaneous tendency of the users under the condition of no query.
10. The method of claim 9,
the plurality of feature items include a temporal feature item, the temporal feature item being converted into a temporal feature item hash function corresponding to a feature value of each user intention; and/or the presence of a gas in the gas,
the plurality of feature items include category feature items, the category feature items corresponding to feature values of each user intention are converted into a category feature item hash function; and/or the presence of a gas in the gas,
the plurality of feature items includes a time-cumulative feature item, and the time-cumulative feature item is converted into a time-cumulative feature item hash function corresponding to a feature value of each user intention.
11. The method according to claim 10, wherein the step of performing model training according to the feature value and the user historical behavior data to obtain the weight value of each feature item comprises:
and performing model training on a time characteristic item hash function, a category characteristic item hash function and/or an aging-accumulation characteristic item hash function according to the historical behavior data of the user to obtain the weight value of each characteristic item.
12. The method of claim 9, further comprising, after the step of obtaining the current probability distribution of the plurality of user intentions:
and identifying the user intention corresponding to the maximum probability value in the probability distribution as a target user intention.
13. An electronic device, comprising:
a processor; and the number of the first and second groups,
a memory for storing executable instructions of the processor;
wherein the processor is configured to perform the steps of the identification method of any one of claims 9-12 by executing the executable instructions.
14. A storage medium on which a computer program is stored which, when being executed by a processor, carries out the steps of the identification method according to any one of claims 9 to 12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711219349.4A CN108230007B (en) | 2017-11-28 | 2017-11-28 | User intention identification method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711219349.4A CN108230007B (en) | 2017-11-28 | 2017-11-28 | User intention identification method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108230007A CN108230007A (en) | 2018-06-29 |
| CN108230007B true CN108230007B (en) | 2021-09-10 |
Family
ID=62653016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711219349.4A Active CN108230007B (en) | 2017-11-28 | 2017-11-28 | User intention identification method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108230007B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110765346B (en) * | 2018-07-10 | 2023-04-18 | 阿里巴巴集团控股有限公司 | User intention mining method, device and equipment |
| CN109087135B (en) * | 2018-07-25 | 2020-08-28 | 百度在线网络技术（北京）有限公司 | Mining method and device for user intention, computer equipment and readable medium |
| CN109299994B (en) * | 2018-07-27 | 2021-09-24 | 北京三快在线科技有限公司 | Recommendation method, device, equipment and readable storage medium |
| CN109492104B (en) * | 2018-11-09 | 2023-01-31 | 北京汇钧科技有限公司 | Training method, classification method, system, device and medium of intention classification model |
| CN109582865A (en) * | 2018-11-19 | 2019-04-05 | 北京奇虎科技有限公司 | A kind of method and device of pushing application program |
| CN109710941A (en) * | 2018-12-29 | 2019-05-03 | 上海点融信息科技有限责任公司 | User's intension recognizing method and device based on artificial intelligence |
| CN110162535B (en) * | 2019-03-26 | 2023-11-07 | 腾讯科技（深圳）有限公司 | Search method, apparatus, device and storage medium for performing personalization |
| US11216832B2 (en) | 2019-06-24 | 2022-01-04 | Advanced New Technologies Co., Ltd. | Predicting future user transactions |
| CN110363571B (en) * | 2019-06-24 | 2020-08-28 | 阿里巴巴集团控股有限公司 | Transaction user advance perception method and device |
| CN111177571B (en) * | 2020-01-13 | 2020-11-27 | 海南大学 | Value-driven dynamic recommendation system with multi-factor dimension space and multi-mesoscale fusion |
| CN111400473A (en) * | 2020-03-18 | 2020-07-10 | 北京三快在线科技有限公司 | Method and device for training intention recognition model, storage medium and electronic equipment |
| CN112711717A (en) * | 2021-03-26 | 2021-04-27 | 北京三快在线科技有限公司 | Room product searching method and device and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105468161A (en) * | 2016-01-21 | 2016-04-06 | 北京百度网讯科技有限公司 | Instruction execution method and device |
| CN105488221A (en) * | 2015-12-25 | 2016-04-13 | 北京奇虎科技有限公司 | Method and system for recommending query terms for conducting searching in search interface |
| US20160140125A1 (en) * | 2014-11-18 | 2016-05-19 | Yahoo! Inc. | Method and system for providing query suggestions based on user feedback |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779230A (en) * | 2016-12-27 | 2017-05-31 | 北京三快在线科技有限公司 | Order volume predictor method and device, electronic equipment |
| CN107146089B (en) * | 2017-03-29 | 2020-11-13 | 北京三快在线科技有限公司 | Method and device for identifying bill swiping and electronic equipment |
-
2017
- 2017-11-28 CN CN201711219349.4A patent/CN108230007B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160140125A1 (en) * | 2014-11-18 | 2016-05-19 | Yahoo! Inc. | Method and system for providing query suggestions based on user feedback |
| CN105488221A (en) * | 2015-12-25 | 2016-04-13 | 北京奇虎科技有限公司 | Method and system for recommending query terms for conducting searching in search interface |
| CN105468161A (en) * | 2016-01-21 | 2016-04-06 | 北京百度网讯科技有限公司 | Instruction execution method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108230007A (en) | 2018-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230007B (en) | User intention identification method and device, electronic equipment and storage medium | |
| CN107424043B (en) | Product recommendation method and device and electronic equipment | |
| CN110163647B (en) | Data processing method and device | |
| CN107341268B (en) | Hot searching ranking method and system | |
| AU2006332534B2 (en) | Predicting ad quality | |
| CN110175895B (en) | Article recommendation method and device | |
| WO2019233077A1 (en) | Ranking of business object | |
| CN110555753A (en) | recommendation-based ranking control method and device, computer equipment and storage medium | |
| CN111695084A (en) | Model generation method, credit score generation method, device, equipment and storage medium | |
| CN111523964A (en) | Clustering-based recall method and apparatus, electronic device and readable storage medium | |
| CN110598126B (en) | Cross-social network user identity recognition method based on behavior habits | |
| CN106997371B (en) | Method for constructing single-user intelligent map | |
| CN113392920B (en) | Method, apparatus, device, medium, and program product for generating cheating prediction model | |
| CN111061948A (en) | User label recommendation method and device, computer equipment and storage medium | |
| CN112651790B (en) | OCPX self-adaptive learning method and system based on user touch in quick-elimination industry | |
| CN117540336A (en) | Time sequence prediction method and device and electronic equipment | |
| CN110162535B (en) | Search method, apparatus, device and storage medium for performing personalization | |
| CN112115354A (en) | Information processing method, information processing apparatus, server, and storage medium | |
| CN113204699A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN116541592A (en) | Vector generation method, information recommendation method, device, equipment and medium | |
| EP3493082A1 (en) | A method of exploring databases of time-stamped data in order to discover dependencies between the data and predict future trends | |
| CN114782062A (en) | Commodity recall optimization method and device, equipment, medium and product thereof | |
| CN112711717A (en) | Room product searching method and device and electronic equipment | |
| CN116484085A (en) | Information delivery method, device, equipment, storage medium and program product | |
| CN117333258A (en) | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium |
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 |