CN111522920A - Method and related device for dynamically recommending initial words in intelligent customer service - Google Patents
Method and related device for dynamically recommending initial words in intelligent customer service Download PDFInfo
- Publication number
- CN111522920A CN111522920A CN201910775211.5A CN201910775211A CN111522920A CN 111522920 A CN111522920 A CN 111522920A CN 201910775211 A CN201910775211 A CN 201910775211A CN 111522920 A CN111522920 A CN 111522920A
- Authority
- CN
- China
- Prior art keywords
- candidate set
- initial language
- initial
- language candidate
- historical usage
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000010606 normalization Methods 0.000 claims description 6
- 230000003068 static effect Effects 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Human Computer Interaction (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a method for recommending a beginning language of a customer service system, which comprises the following steps: acquiring the historical usage amount of each initial language candidate set; then acquiring the real-time exposure number and the click number of each initial language candidate set; scoring each initial language candidate set by using the historical usage amount of each initial language candidate set, the real-time exposure number of each initial language candidate set and the click number of each initial language candidate set; and recommending the initial language candidate set according to the scores of the initial language candidate sets. Through the mode, the initial words recommended by the method not only can comprise historical record conditions, but also can be combined with real-time clicking conditions, so that the static recommending mode of initial word recommendation is changed into the dynamic recommending mode, each user can be guaranteed to see different recommending information when clicking an intelligent customer service chat page every time, timeliness is achieved, and user experience is improved.
Description
Technical Field
The invention relates to the technical field of intelligent customer service systems, in particular to a method and a related device for dynamically recommending a beginning language in intelligent customer service.
Background
The intelligent customer service product is a product of AI, and the intelligent customer service product is widely applied to the industries of e-commerce, finance, medical treatment and the like. The intelligent customer service can provide service continuously for 24 hours, thereby improving the efficiency of enterprises or reducing the labor cost for the enterprises.
The beginning is an important function in intelligent customer service, and many enterprises are also called as 'question and answer before'. Specifically, the beginning language function means that the client enters the intelligent client service chat page and does not speak, and at the moment, the intelligent client service provides information recommended to the user.
At present, the initial words recommended by the intelligent customer service are generally counted based on historical data, for example, chat log data of set days is counted, and after a user enters a question-answering interface, a recommended word with a high historical click frequency is recommended to the user. That is, in the prior art, the recommended information in the initial language recommended to the user is static, the recommendation results seen by all users are the same, and the recommendation results are relatively old. Therefore, the recommendation of the conventional intelligent customer service beginning speech function is not real-time, and is difficult to meet the sudden user demand.
Disclosure of Invention
The invention provides a method and a related device for recommending a beginning language of a customer service system, which are used for solving the problems that the intelligent customer service beginning language in the prior art can only provide the same static recommendation information for all users and the recommendation information is old and single.
In order to solve the above technical problem, the present invention provides a method for recommending a beginning phrase of a customer service system, the method comprising: acquiring the historical usage amount of each initial language candidate set; acquiring the real-time exposure number and the click number of each initial language candidate set; scoring each initial language candidate set by using the historical usage amount of each initial language candidate set, the real-time exposure number and the click number of each initial language candidate set; and recommending the initial language candidate set according to the scores of the initial language candidate sets.
Wherein the step of scoring each initial language candidate set by using the historical usage amount of each initial language candidate set and the real-time exposure number and click number of each initial language candidate set comprises:
determining the current click rate of each initial language candidate set according to the click number and the exposure number of each initial language candidate set;
scoring each initial language candidate set according to the current click rate and the historical usage amount of each initial language candidate set;
the step of scoring each initial language candidate set according to the current click rate and the historical usage amount of each initial language candidate set specifically includes: each candidate set of initials is scored according to the following equation (1):
score(si)＝(r1，r2，r3，…ri，…,rn)＝alpha*ki+beta*ci (1)，
wherein si is the ith initial word candidate set, ri is the score of the ith initial word candidate set, n is the number of initial words, n is a positive integer, and ki is the historical usage amount of the ith initial word candidate set after normalization processing; ci is the current click rate of the ith initial language candidate set; alpha is a weight coefficient of the historical usage amount of the ith initial language candidate set; beta is a weight coefficient of the current click rate of the ith initial language candidate set, and alpha + beta is 1;
the step of acquiring the historical usage amount of each initial language candidate set comprises the following steps: performing the normalization processing on the historical usage amount of each initial language candidate set;
the step of normalizing the historical use amount of each initial language candidate set comprises the following steps: the historical usage of each candidate set of initials is normalized by the following equation (2):
ki＝Hi/(max(H)–min(H)) (2)，
wherein, Hi is the historical usage of the ith initial language candidate set, max (h) is the maximum historical usage of the initial language candidate set, and min (h) is the minimum historical usage of the initial language candidate set;
the step of recommending the initial words according to the scores of the initial word candidate sets specifically comprises the following steps: determining the selection times of the initial language candidate sets in a preset total time according to the scores of the initial language candidate sets and a preset rule, and determining the selection probability of the initial language candidate sets according to the ratio of the corresponding selection times to the preset total time; and selecting a certain number of initial language candidate sets according to the selected probability of each initial language candidate set to perform recommended exposure.
Wherein the preset rule comprises:
when i is not equal to 1, calculating an upper bound ai and a lower bound bi of the interval length Y [ ai, bi) corresponding to the selected times of the ith initial language candidate set by the following formulas (3) and (4):
ai＝[sum(r1，r2，..，r(i-1))/sum(r1，r2，r3，…，rn)]*L (3)
bi＝[sum(r1，r2，..，r(i-1)，ri)/sum(r1，r2，r3，…，rn)]*L (4)
when i is equal to 1, setting the upper bound ai of the interval length Y corresponding to the selected times of the ith initial language candidate set to 0, and calculating the lower bound bi of the interval length corresponding to the selected times of the initial language candidate set through the formula (4); and L is a preset total length corresponding to the preset total times.
In order to solve the technical problem, the invention also provides a recommendation device for the initial words of the customer service system, which comprises a history acquisition module, a real-time acquisition module, a scoring module and a recommendation module;
the history acquisition module is used for acquiring the history usage amount of each initial language candidate set; the real-time acquisition module is used for acquiring the real-time exposure number and the click number of each initial language candidate set; the scoring module is used for scoring each initial language candidate set by utilizing the historical usage of each initial language candidate set and the real-time exposure number and click number of each initial language candidate set; and the recommending module is used for recommending the initial words according to the scores of the initial word candidate sets.
In order to solve the above technical problem, the present invention further provides a customer service system, including: a processor and a memory, wherein the memory stores the historical data and real-time data of the user, and the processor is used for executing the recommendation method of any one of the above items.
In order to solve the above technical problem, the present invention further provides a storage device, which stores program data that can be executed to implement any one of the recommendation methods described above.
The invention has the beneficial effects that: different from the prior art, the initial language recommendation method and the system generate the initial language candidate set by simultaneously acquiring the historical data and the real-time data of the user, generate and push the initial language recommendation information to the user by the initial language candidate set scoring evaluation mode and the initial language candidate set recommendation generation mode of the invention, and change the initial language recommendation from a static recommendation mode to a dynamic recommendation mode by combining the historical record and the real-time click condition, so that the recommendation information seen by each user clicking an intelligent customer service chat page every time is different, the timeliness is realized, and the user experience is improved.
Drawings
FIG. 1 is a flowchart illustrating a method for recommending a beginning phrase according to an embodiment of the present invention;
FIG. 2 is a schematic sub-flow chart of step S13 in the embodiment shown in FIG. 1;
FIG. 3 is a schematic sub-flow chart of step S14 in the embodiment shown in FIG. 1;
FIG. 4 is a schematic structural diagram of an embodiment of a device for recommending beginning words according to the present invention;
FIG. 5 is a schematic diagram of an embodiment of a customer service system provided by the present invention;
fig. 6 is a schematic structural diagram of an embodiment of a memory device provided in the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, fig. 1 is a flowchart illustrating a method for recommending beginnings according to an embodiment of the present invention, where the method for recommending beginnings includes the following steps:
s11: and acquiring the historical usage amount of each initial language candidate set.
The initial language generally includes two ways, one is labeled, i.e., one label for each question, each label may correspond to multiple questions, as described in table 1 below, and the other is question-and-answer, i.e., one question corresponds to one answer, as shown in table 2.
TABLE 1 tagged question-answer schema example
TABLE 2 question-answering mode example
| Method of inquiry | Answer to the question |
| How may interest be reduced? | Interest is uniform and cannot be changed, thanks! |
| What activity was there recently? | If there are more activities, please click the website to view. |
In the present embodiment, the historical usage amount of each candidate set of initial words, that is, the usage amount of each candidate set of initial words in a historical preset time period, is collected first. Specifically, the usage amount of the initial language in the chat records of all users and the smart customer service within the set time period is acquired, and in an optional embodiment, the usage amount of the initial language candidate set refers to the click amount of the initial language candidate set, a question input by the user through characters, voice, body feeling or other embodiments, and the like. The initial language candidate set includes the question-answer initial language candidate set and a tag-type initial language candidate set, where one tag-type initial language candidate set corresponds to only one tag, and although one tag may correspond to a plurality of questions, when a certain tag-type initial language candidate set is selected, it only presents one relevant question to the user, and when a question-answer initial language candidate set also corresponds to only one question, only one relevant question is presented when a certain question-answer initial language candidate set is recommended.
In a specific embodiment, the historical usage amount can be obtained by a packet sniffer mode. (the packet sniffer is used according to the principle that a request sent by a visitor passes through the packet sniffer before reaching the server, and then the packet sniffer sends the request to the server, the data collected by the packet sniffer is processed by the processing server and then stored in the database.)
The set time period may be set as needed, for example, the time may be set according to the problem heat degree, such as 5 days, 7 days, 10 days, 15 days, 30 days, or other time. Preferably, 7 days are selected as the set period of time.
It should be noted that some initial word candidate sets may not be clicked or may be clicked by a small amount within the set time period, but in order to improve the accuracy of the subsequent scoring, the usage amount of the initial word candidate sets may be set to 0, instead of being counted.
For example, if the total number of the candidate sets of the beginning words is n, the candidate set of the beginning words is set to si, and the set of the candidate sets of the beginning words is set to S, that is, S ═ S (S1, S2, S3 …, sn); setting the historical usage of the corresponding initial language candidate set as H, and setting the historical usage of the candidate set as: h is (H1, H2, H3, …, hn) the usage amount of the candidate set of beginning words, and hi is the usage amount of the i-th candidate set of beginning words in the history preset time period. For example, the usage amount of each candidate set of initial words in 7 days is counted, the candidate sets of initial words are 10, the set of corresponding candidate sets of initial words is S ═ S (S1, S2, S3, S4, S5, S6, S7, S8, S9, S10), and the usage amount of each acquired candidate set of initial words in 7 days of the history preset period is:
H＝(h1，h2，h3，h4，h5，h6，h7，h8，h9，h10)。
s12: and acquiring the real-time exposure number and the click number of each initial language candidate set.
In a specific embodiment, the exposure number and the click number can be obtained by a packet sniffer manner as well.
If the candidate set of initials is a tagged initials, the number of exposures is the number of questions to be pushed to the user, and if the candidate set of initials is a question-and-answer initials, the number of exposures is also the number of questions to be pushed to the user, for example: if 3 tag-type initial language candidate sets and 2 question-answer-type initial language candidate sets are selected for recommendation, since both candidate sets recommend only one corresponding question, the exposure number is the same as the number of recommended initial language candidate sets, and is 5.
There are generally two ways to count clicks, one is cumulative clicks, such as: the user clicks 3 pieces of information in the recommendation information (the same information is clicked for multiple times for 1 time), and then 3 clicks are calculated; one is a non-additive click, such as: the intelligent customer service recommends once, and the user only counts the number of clicks to be 1 no matter how many times the user clicks; and if there is no click behavior, the statistical number of clicks is 0.
S13: and scoring each initial language candidate set by using the historical usage amount of each initial language candidate set and the real-time exposure number and click number of each initial language candidate set.
Alternatively, the exposure number of each candidate set of beginning words and the corresponding click number can calculate the click rate of the corresponding candidate set of beginning words, and the specific steps are as shown in fig. 2, and step S131 includes:
s131: and determining the current click rate of each initial language candidate set according to the click number and the exposure number of each initial language candidate set.
Specifically, the click-through rate calculation rule is as follows:
setting the exposure number of the initial language candidate set as P, and the exposure number corresponding to the candidate set as follows: p ═ P (P1, P2, P3, …, pn), pi is the exposure number of the ith startword candidate set; setting the number of clicks corresponding to the initial language candidate set as Q, and the number of clicks corresponding to the candidate set as follows: q ═ Q1, Q2, Q3, …, qn, qi is the number of clicks of the ith candidate set of initials, the click rate corresponding to the candidate set of initials is set as C, and the click rate corresponding to the candidate set is: c ═ C1, C2, C3, …, cn, ci is the click rate of the ith startword candidate set, and the click rate calculation formula is:
ci＝qi/pi。
optionally, the step S132 includes, as shown in fig. 2, scoring the historical usage amount through the normalization process and the click rate of the corresponding candidate set of the initial language:
s132: and normalizing the historical use amount of each initial language candidate set.
Since the historical usage and click rate of each candidate set of initials are not in one dimension, for example: the historical usage amount may be 100, 5000, etc., but the click rate ranges from 0 to 1, the historical usage amount needs to be normalized, and a maximum and minimum normalization method is adopted in consideration of the fact that the historical usage amount is fixed.
Let n be the total number of candidate sets, si be the initial language candidate set, and S be the set of initial language candidate sets, i.e., S is (S1, S2, S3 …, sn). Setting the historical usage of the corresponding initial language candidate set as H, and setting the historical usage of the candidate set as: h is (H1, H2, H3, …, hn) the usage amount of the candidate set of the beginning words, hi is the usage amount of the i-th candidate set of the beginning words in a historical preset time period;
the historical usage is normalized to: hi/(max (h) -min (h)).
S133: and scoring each initial language candidate set according to the current click rate and the historical usage amount of each initial language candidate set.
The score calculation formula for the candidate set si of the beginning is as follows:
score(si)＝(r1，r2，r3，…ri，…,rn)＝alpha*ki+beta*ci＝ri，
the two parameters of alpha and beta are weight super parameters, which satisfy that alpha + beta is 1, and the two weight super parameters of alpha and beta can be adjusted according to actual use, for example, if the user belongs to conservative assignment, the historical data of weight bias can be set to be 0.8 for alpha and 0.2 for beta; if the user belongs to the radical group, the real-time data is weighted, alpha can be set to be 0.3, and beta can be set to be 0.7, but not limited to this.
S14: and recommending the initial language candidate set according to the scores of the initial language candidate sets.
In an alternative embodiment, after the scores of the initial language candidate sets are obtained, a certain number of initial language candidate sets can be recommended directly according to the scores. By means of combination of historical records and real-time clicking conditions, initial language recommendation is changed from a static recommendation mode to a dynamic recommendation mode, the fact that recommendation information seen by each user clicking an intelligent customer service chat page every time is different is guaranteed, timeliness is achieved, and user experience is improved.
Further, when the scores of the candidate sets of the initial words are calculated, recommendation of the candidate sets can be theoretically performed, but since the scores are fixed values, only the candidate set with a high score is recommended during recommendation, so that recommendation information received by a plurality of users is the same, for example, if two candidate sets exist, a candidate set A with a score of 40 and a candidate set B with a score of 60 are recommended, and when only one candidate set is recommended according to the scores, no matter how many users enter an interface, the candidate sets are the same and the candidate sets B with the scores of 60 are all too detailed, and the recommendation method is too detailed; if the probability corresponding to the candidate set is calculated according to the score, in this example, the recommendation probability of the candidate set a is 40%, the probability of the candidate set B is 60%, and if 100 users enter the interface, in an ideal case, 60 users see the candidate set B, and another 40 users see the candidate set a, and so on, so the recommendation method has diversity.
Therefore, in order to further improve the diversity of the recommendation information, in another alternative embodiment, the hit probability may be calculated according to the score of each candidate set of beginning words. Specifically, as shown in fig. 3, fig. 3 is a schematic flow chart of another 3 embodiments of step S141. The method comprises the following steps:
s141: and determining the selection times of the initial language candidate sets in the preset total times according to the scores of the initial language candidate sets and a preset rule, and determining the selection probability of each initial language candidate set according to the ratio of the corresponding selection times to the preset total times.
The logic is as follows: setting the preset total length corresponding to the preset total times as a very large and fixed positive integer L, meanwhile, setting the ith initial word candidate set as si, the probability of selecting si as T, the number of the initial word candidate sets as i, the interval length corresponding to the selected times of the ith initial word candidate set as Y, ai as the lower bound of the interval length, and bi as the upper bound of the interval length.
When i is not equal to 1 then,
let ai ═ L [ sum (r1, r 2., r (i-1))/sum (r1, r2, r3, …, rn) ],
bi＝[sum(r1，r2，..，r(i-1)，ri)/sum(r1，r2，r3，…，rn)]*L，
y belongs to [ ai, bi ],
in terms of the values: y-bi-ai ═ L [ ri/sum (r1, r2, r3, …, rn) ],
si is chosen as probability T ═ bi-ai)/L.
When i is equal to 1, the data is transmitted,
let ai be 0 and make ai be 0,
let bi ═ L [ sum (r1, r 2., r (i-1), ri)/sum (r1, r2, r3, …, rn) ],
y belongs to [0, bi ],
si is chosen as the probability T-bi/L.
That is, combining all the cases that i is not equal to 1 and i is equal to 1, the union of the interval length sets corresponding to all the candidate sets for the initial language is [0, L ], and the interval length corresponding to the number of hits in each candidate set is determined by the score value of the interval length set.
For example, if there are 4 initial candidate sets of a, b, c, and d, the scores are 1 point for a, 4 points for b, 5 points for c, and 10 points for d, respectively, and the total length corresponding to the preset total number of times is set to 100.
When calculating the selection probability of the first candidate set that is the first candidate set,
ai＝0，bi＝[1/(1+4+5+10)]*100＝5，
that is, at this time, the interval length Y of the nail belongs to [0, 5 ],
calculating the selection probability of the second candidate set B candidate set:
ai＝[1/(1+4+5+10)]*100＝5，
bi＝[(1+4)/(1+4+5+10)]*100＝25，
that is, the interval length Y of B at this time belongs to [5, 25 ],
when calculating the selection probability of the third candidate set, i.e. the third candidate set,
ai＝[(1+4)/(1+4+5+10)]*100＝25，
bi＝[(1+4+5)/(1+4+5+10)]*100＝50，
that is, the interval length Y of B at this time belongs to [25, 50 ],
when calculating the selection probability of the fourth candidate set that is the candidate set,
ai＝[(1+4+5)/(1+4+5+10)]*100＝50，
bi＝[(1+4+5+10)/(1+4+5+10)]*100＝100，
i.e., the interval length Y of this time T belongs to [50, 100 ],
in this case, T is 5 ═ 5-0)/100 ═ 5%, T is 20 ═ 25-5)/100 ═ 20%, T is 25 ═ 50-25)/100 ═ 25%, and T is 50 ═ 100-50/100 ═ 50%.
It can be seen that the union of the interval length sets corresponding to the selection times of all the candidate sets is the preset total length corresponding to the preset total length, and the interval length corresponding to the selection times of the candidate sets is determined by the corresponding score.
That is, the larger the number of hits is, the longer the corresponding section length is, the larger the corresponding probability of being selected is, and the smaller the number of hits is, the shorter the corresponding section length is, the smaller the probability of being selected is.
S142: and selecting a certain number of initial language candidate sets according to the selected probability of each initial language candidate set to perform recommended exposure.
After the selection probability of each initial language candidate set is calculated, the initial language candidate sets are selected from all the initial language candidate sets in a non-repeated mode according to the selection probability until a preset number of candidate sets or no candidate sets are selected.
The set number of the initial words can be adjusted, for example, it can be set according to actual needs, such as 3, 5, 7 or other numbers, preferably, 5 are selected as the set number.
By the method, the historical usage amount of each initial language candidate set is obtained, the real-time exposure number and the click number of each initial language candidate set are obtained, the click rate is calculated, then the score is calculated according to the score calculation formula, finally the corresponding interval length generated by the corresponding candidate set is calculated according to the score result, and the corresponding initial language candidate set is recommended according to the probability corresponding to the interval length, so that the limitation that only historical data is considered in the prior art is solved, the sudden user requirement is considered in combination with the historical record and the real-time click, a more time-effective initial language function is brought to the user, and the user experience is improved.
Referring to fig. 4, fig. 4 is a schematic structural diagram of an embodiment of a beginnings recommendation device according to the present invention, and the beginnings recommendation device of the present embodiment includes a history obtaining module 31, a real-time obtaining module 32, a scoring module 33, and a recommendation module 34.
The history obtaining module 31 is configured to obtain a history usage amount of each candidate set of beginning words; this module acquires historical use amount through the package sniffer mode, and wherein, historical preset time can set for according to actual need.
The real-time obtaining module 32 is configured to obtain real-time exposure numbers and click numbers of each initial language candidate set; the module can also acquire the exposure number and the click number in a packet sniffer mode, and simultaneously determine the current click rate of each initial language candidate set according to the click number and the exposure number of each initial language candidate set, wherein the specific click rate calculation formula is as follows: ci-qi/pi.
The scoring module 33 is configured to score each initial language candidate set according to the scoring rule by using the historical usage amount of each initial language candidate set, and the real-time exposure number and click number of each initial language candidate set; firstly, the historical use amount of each initial language candidate set is normalized, and the logic is as follows: and scoring each initial word candidate set by the current click rate and the historical usage of each initial word candidate set, wherein in a specific embodiment, the score calculation formula is score (si) ═ alpha ═ ki + beta ═ ci ═ ri.
The recommending module 34 is configured to recommend the initial words according to the selected probabilities corresponding to the scores of the initial word candidate sets; determining the selection times of each initial language candidate set in the preset total times according to a preset rule, and determining the selection probability of each initial language candidate set according to the ratio of the corresponding selection times to the total times; and finally, determining a set number of initial words according to the selected probability of each initial word candidate set to perform recommended exposure.
Based on the same inventive concept, the present invention further provides a customer service system, which can be executed to implement the method for recommending the beginning words of any of the above embodiments, please refer to fig. 5, fig. 5 is a schematic structural diagram of an embodiment of the customer service system provided by the present invention, and the customer service system includes a processor 41 and a memory 42.
The memory 42 is used for storing the usage amount of each initial phrase and the exposure number and the click number of each initial phrase in the historical preset time of the user acquired by the packet sniffer mode.
The processor 41 is configured to obtain a click rate of the candidate set of corresponding beginning words by ci ═ qi/pi algorithm according to the exposure number and the click rate of each beginning word; and normalizing the historical use amount of each initial language candidate set according to a ki ═ Hi/(max (H) -min (H)) algorithm.
Scoring each initial language candidate set according to the click rate of each initial language candidate set and the historical usage amount of each initial language candidate set, wherein the scoring formula is as follows: score (si) ═(r1, r2, r3, … ri, …, rn) ═ alpha ki + beta ci ═ ri.
And finally, determining the selection times of the initial language candidate sets in the preset total times according to the scores of the initial language candidate sets and a preset rule, determining the selection probability of each initial language candidate set according to the ratio of the corresponding selection times to the total times, and determining the set number of initial language candidate sets according to the selection probability of each initial language candidate set to carry out recommended exposure.
Based on the same inventive concept, the present invention further provides a memory device, please refer to fig. 6, and fig. 6 is a schematic structural diagram of an embodiment of the memory device according to the present invention. Program data 51 is stored in the storage device 50, and the program data 51 may be a program or a command, and the program data is capable of executing the acquisition of the historical usage amount of each candidate set of beginning words; acquiring the real-time exposure number and the click number of each initial language candidate set; scoring each initial language candidate set by using the historical usage amount of each initial language candidate set and the real-time exposure number and click number of each initial language candidate set; and recommending the initial language candidate set according to the selection probability corresponding to the initial language candidate set score.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A method for recommending a beginning phrase of a customer service system, the method comprising:
acquiring the historical usage amount of each initial language candidate set;
acquiring the real-time exposure number and the click number of each initial language candidate set;
scoring each initial language candidate set by using the historical usage amount of each initial language candidate set, the real-time exposure number and the click number of each initial language candidate set;
and recommending the initial language candidate set according to the scores of the initial language candidate sets.
2. The method for recommending slogans in a customer service system according to claim 1, wherein the step of scoring each of the candidate setinguals using the historical usage amount of each of the candidate setinguals and the exposure count and the click count of each of the candidate setinguals in real time comprises:
determining the current click rate of each initial language candidate set according to the click number and the exposure number of each initial language candidate set;
and scoring each initial language candidate set according to the current click rate and the historical usage amount of each initial language candidate set.
3. The method for recommending slogans in a customer service system according to claim 2, wherein the step of scoring each of the candidate sets of slogans by the current click rate and the historical usage amount of each of the candidate sets of slogans specifically includes:
scoring said initial word candidate sets according to the following equation (1):
score(si)＝(r1，r2，r3，…ri，…,rn)＝alpha*ki+beta*ci (1)，
wherein si is the ith initial word candidate set, ri is the score of the ith initial word candidate set, n is the number of initial words, n is a positive integer, and ki is the historical usage amount of the ith initial word candidate set after normalization processing; ci is the current click rate of the ith initial language candidate set; alpha is a weight coefficient of the historical usage amount of the ith initial language candidate set; beta is a weight coefficient of the current click rate of the ith initial language candidate set, and alpha + beta is 1.
4. The method of claim 3, further comprising, after obtaining the historical usage of each candidate set of initials:
and performing the normalization processing on the historical usage amount of each initial language candidate set.
5. The method of recommending initials by a customer service system according to claim 4, wherein said step of normalizing the historical usage amount of each of said candidate sets of initials comprises:
normalizing the historical usage of each of the initial language candidate sets by the following formula (2):
ki＝Hi/(max(H)–min(H)) (2)，
where, Hi is the historical usage of the ith candidate set for the beginning, max (h) is the maximum historical usage of the candidate set for the beginning, and min (h) is the minimum historical usage of the candidate set for the beginning.
6. The method for recommending starting words in a customer service system according to any one of claims 3 to 5, wherein the step of recommending the candidate set of starting words based on the score of each candidate set of starting words specifically comprises:
determining the selection times of the initial language candidate sets in a preset total time according to the scores of the initial language candidate sets and a preset rule, and determining the selection probability of the initial language candidate sets according to the ratio of the corresponding selection times to the total time;
and selecting a certain number of initial language candidate sets according to the selected probability of each initial language candidate set to perform recommended exposure.
7. The method for recommending the beginning words of a customer service system according to claim 6, wherein said preset rules comprise:
when i is not equal to 1, calculating an upper bound ai and a lower bound bi of the interval length Y [ ai, bi) corresponding to the selected times of the ith initial language candidate set by the following formulas (3) and (4):
ai＝[sum(r1，r2，..，r(i-1))/sum(r1，r2，r3，…，rn)]*L (3)
bi＝[sum(r1，r2，..，r(i-1)，ri)/sum(r1，r2，r3，…，rn)]*L (4)
when i is equal to 1, setting the upper bound ai of the interval length Y corresponding to the selected times of the ith initial language candidate set to 0, and calculating the lower bound bi of the interval length corresponding to the selected times of the initial language candidate set through the formula (4); and L is a preset total length corresponding to the preset total times.
8. A recommendation device for a beginning language of a customer service system is characterized by comprising a history acquisition module, a real-time acquisition module, a scoring module and a recommendation module,
the history acquisition module is used for acquiring the history usage amount of each initial language candidate set;
the real-time acquisition module is used for acquiring the real-time exposure number and the click number of each initial language candidate set;
the scoring module is used for scoring the initial language candidate sets by utilizing the historical usage amount of the initial language candidate sets, the real-time exposure number and click number of the initial language candidate sets;
and the recommending module is used for recommending the initial language candidate sets according to the scores of the initial language candidate sets.
9. A customer service system, characterized in that the customer service system comprises: a processor and a memory, said memory having stored therein said historical data and real-time data of said user, said processor being adapted to perform the recommendation method of any one of claims 1-8.
10. A storage device, characterized in that the storage device stores program data executable to implement the recommendation method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910775211.5A CN111522920B (en) | 2019-08-21 | 2019-08-21 | Method and related device for dynamically recommending initial words in intelligent customer service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910775211.5A CN111522920B (en) | 2019-08-21 | 2019-08-21 | Method and related device for dynamically recommending initial words in intelligent customer service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111522920A true CN111522920A (en) | 2020-08-11 |
| CN111522920B CN111522920B (en) | 2021-12-03 |
Family
ID=71900356
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910775211.5A Active CN111522920B (en) | 2019-08-21 | 2019-08-21 | Method and related device for dynamically recommending initial words in intelligent customer service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111522920B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101079039A (en) * | 2006-06-30 | 2007-11-28 | 腾讯科技（深圳）有限公司 | Search advertisement sequencing system and method |
| CN103246985A (en) * | 2013-04-26 | 2013-08-14 | 北京亿赞普网络技术有限公司 | Advertisement click rate predicting method and device |
| CN104424291A (en) * | 2013-09-02 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for sorting search results |
| CN105654326A (en) * | 2014-11-14 | 2016-06-08 | 阿里巴巴集团控股有限公司 | Information processing system and information processing method |
| CN106682925A (en) * | 2015-11-06 | 2017-05-17 | 北京奇虎科技有限公司 | Method and device for recommending advertisement content |
-
2019
- 2019-08-21 CN CN201910775211.5A patent/CN111522920B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101079039A (en) * | 2006-06-30 | 2007-11-28 | 腾讯科技（深圳）有限公司 | Search advertisement sequencing system and method |
| CN103246985A (en) * | 2013-04-26 | 2013-08-14 | 北京亿赞普网络技术有限公司 | Advertisement click rate predicting method and device |
| CN104424291A (en) * | 2013-09-02 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for sorting search results |
| CN105654326A (en) * | 2014-11-14 | 2016-06-08 | 阿里巴巴集团控股有限公司 | Information processing system and information processing method |
| CN106682925A (en) * | 2015-11-06 | 2017-05-17 | 北京奇虎科技有限公司 | Method and device for recommending advertisement content |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111522920B (en) | 2021-12-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7406434B1 (en) | System and method for improving the performance of electronic media advertising campaigns through multi-attribute analysis and optimization | |
| Bauwens et al. | The stochastic conditional duration model: a latent variable model for the analysis of financial durations | |
| US8121886B2 (en) | Confidence based selection for survey sampling | |
| CN107122467B (en) | Search engine retrieval result evaluation method and device and computer readable medium | |
| US20110191170A1 (en) | Similarity function in online advertising bid optimization | |
| US20110191169A1 (en) | Kalman filter modeling in online advertising bid optimization | |
| KR20080114761A (en) | A behavioral targeting system | |
| US11610580B2 (en) | System and method for determining reasons for anomalies using cross entropy ranking of textual items | |
| EP3358473A1 (en) | Resource combination processing method, apparatus, equipment and computer storage medium | |
| US20110301926A1 (en) | Method or system to evaluate strategy decisions | |
| CN103177371B (en) | A kind of information displaying method and device | |
| Malam et al. | The determinants of long run economic growth in Nigeria | |
| Shita et al. | The impact of agricultural technology adoption on income inequality: a propensity score matching analysis for rural Ethiopia | |
| CN116188140A (en) | Credit financial risk policy production method and device | |
| CN111061948A (en) | User label recommendation method and device, computer equipment and storage medium | |
| Mohri et al. | Non-parametric revenue optimization for generalized second price auctions | |
| CN111522920B (en) | Method and related device for dynamically recommending initial words in intelligent customer service | |
| Cappelli et al. | Change point analysis of imprecise time series | |
| Almus | Job creation through public start-up assistance? | |
| Koustas et al. | Evidence of nonlinear mean reversion in the real interest rate | |
| Choy | Trade cycles in a re-export economy: the case of Singapore | |
| Tan et al. | Investment recommendation with total capital value maximization in online P2P lending | |
| CN111209484A (en) | Product data pushing method, device, equipment and medium based on big data | |
| Ouiminga | The effects of adopting sustainable farming practices on smallholders | |
| Van der Sluis et al. | Entrepreneurship Selection and Performance |
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 | ||
| OL01 | Intention to license declared | ||
| OL01 | Intention to license declared |