CN109993382B

CN109993382B - Express delivery order recommending method and system

Info

Publication number: CN109993382B
Application number: CN201711475912.4A
Authority: CN
Inventors: 雷紫霖; 金晶; 杜堃; 王晶; 白文勇
Original assignee: SF Technology Co Ltd
Current assignee: SF Technology Co Ltd
Priority date: 2017-12-29
Filing date: 2017-12-29
Publication date: 2023-05-19
Anticipated expiration: 2037-12-29
Also published as: CN109993382A

Abstract

The invention discloses a method for recommending a delivery order of an express delivery person, which comprises the following steps: s1, acquiring historical dispatch data of an express delivery person; s2, converting historical dispatch sequence data of the courier into a training set of a model, and training a classifier by using the training set to obtain a prediction model; s3, aiming at a dispatch process which hopes to obtain a recommendation dispatch sequence, converting relevant information of the dispatch process into an input vector of a prediction model, and calling the prediction model to obtain a dispatch sequence recommendation list. The invention has the beneficial effects that: according to the express delivery order recommending method and system, the order model is obtained by training the classifier by using historical delivery data of old staff, and the model is completely based on the data without collecting information from the staff in advance. According to the sequence model, a dispatch sequence recommendation list is provided for the courier, so that the work target of the courier is clearer and simpler, and meanwhile, the learning cost and the time cost are saved.

Description

Express delivery order recommending method and system

Technical Field

The invention relates to the technical field of logistics information processing, in particular to a method and a system for recommending delivery order of couriers.

Background

In the prior art, the order of delivery of the express items is manually selected by the express men according to their own business experience, which mainly derives from the familiarity of the express men to the geographical information of the delivery area and the distribution of the amount of the express items in the same day. The disadvantage of relying on manual assignment of tasks is that the friendliness is poor for new staff, who are not familiar with the geographical information of the cell area, and that a certain learning cost is required in the process of planning the dispatch sequence. On the other hand, when manual distribution is relied on, a specific dispatch sequence only exists in the courier, and the data cannot be acquired to provide more accurate aging service for clients.

For the problems in the related art, no effective solution has been proposed at present.

Disclosure of Invention

Aiming at the technical problems in the related art, the invention provides a method and a system for recommending the dispatch sequence of an express delivery person, which restore the dispatch sequence with high efficiency according to the historical dispatch data of old staff, so that new staff can find the most favorable dispatch sequence more quickly.

In order to achieve the technical purpose, the technical scheme of the invention is realized as follows:

a method for recommending order of delivery of couriers comprises the following steps:

s1, acquiring historical dispatch data of an express delivery person;

s2, converting historical dispatch sequence data of the courier into a training set of a model, and training a classifier by using the training set to obtain a prediction model;

s3, aiming at a dispatch process which hopes to obtain a recommendation dispatch sequence, converting relevant information of the dispatch process into an input vector of a prediction model, and calling the prediction model to obtain a dispatch sequence recommendation list.

Preferably, the historical dispatch data in step S1 includes information about a plurality of dispatch processes, and the information about each dispatch process includes at least a dispatch address and a dispatch sequence related to executing a plurality of dispatch tasks.

Preferably, the training set for converting the historical dispatch sequence data of the courier into a model in step S2 further includes the following steps:

s2.1, selecting one from a plurality of dispatch processes, and converting dispatch addresses related in the dispatch process into input vectors of an interest point judgment model;

s2.2, calling an interest point judgment model to obtain an interest point;

s2.3, arranging the obtained interest points according to the dispatching sequence involved in the dispatching process to obtain an interest point sequence list;

s2.4, extracting an input vector and an output vector from the interest point sequence list as training data;

s2.5, repeating the steps S2.1-S2.4 until all the dispatch processes are converted into training data, namely the training set.

Preferably, in the step S2.1, the input vector of the interest point determination model is a joint vector formed by converting the dispatch address into a keyword vector and longitude and latitude information.

Preferably, in the step S2.4, the input vector is a joint vector formed by the number of pieces of each interest point and the current position of the courier before the courier performs one of the specific dispatch tasks in a specific dispatch process, and the output vector is the interest point to which the courier actually performs the specific dispatch task.

Preferably, the step S3 further includes the steps of:

s3.1, converting current state information when a dispatch process starts to be carried out into an input vector of a prediction model, and calling the prediction model to obtain interest points to which the first step is recommended;

s3.2, updating current state information, converting the updated current state information into an input vector of a prediction model, and calling the prediction model to obtain interest points recommended to go to next step;

s3.3, repeating the step S3.2 to obtain a plurality of interest points, and arranging the interest points according to the acquisition order, namely, a dispatch order recommendation list.

Preferably, the current state information includes at least:

in a dispatch process, before the courier performs one of dispatch tasks, the actual position of the courier; the method comprises the steps of,

in a dispatch process, before the courier performs one of the dispatch tasks, all dispatch addresses related to the remaining couriers carried by the courier.

Preferably, the input vector of the prediction model is a joint vector formed by the quantity of the parts of each interest point before the courier executes one of the dispatch tasks and the current position of the courier in a dispatch process.

Preferably, the classifier is a random forest classifier.

In another aspect of the present invention, there is provided a dispatch sequence recommendation system based on courier history data, including:

an information acquisition unit: the method comprises the steps of acquiring related information of a dispatch process;

an information conversion unit: the input vector is used for converting the related information of the dispatch process into a prediction model;

prediction model: the method comprises the steps of obtaining recommended interest points according to input vectors of a prediction model;

the prediction model is obtained by converting historical dispatch sequence data of couriers into a training set of models and training a classifier by using the training set.

The invention has the beneficial effects that: according to the express delivery order recommending method and system, the order model is obtained by training the classifier by using historical delivery data of old staff, and the model is completely based on the data without collecting information from the staff in advance. According to the sequence model, a dispatch sequence recommendation list is provided for the courier, so that the work target of the courier is clearer and simpler, and meanwhile, the learning cost and the time cost are saved.

By using the method and the system for recommending the dispatch sequence of the couriers, the dispatch sequence of the couriers can be known in advance, so that the time point when each dispatch task is executed can be accurately predicted, and the possibility of providing accurate dispatch time prediction for clients is provided.

In addition, in the method and the system for recommending the dispatch sequence of the couriers, a better sequence model can be obtained by continuously training a classifier by using new data, and favorable conditions are provided for further adding new dispatch sequence rules.

Finally, the method for solving the shortest total path of the dispatch generally has higher requirements on geographic position information, once the data quality of the geographic information is too low, serious errors occur in path prediction, and in the logistics industry, the situation that the data quality of the geographic information is low, such as information deficiency and misprinting of address text data, and the situation that the geographic position information is inaccurate, can obtain accurate geographic position information by using keyword vectors and longitude and latitude information as joint vectors, and can obtain interest points by using accurate geographic position information to call interest point judging models and then convert the interest points into training data to train classifiers, so that the finally obtained sequence model has certain robustness to the problem that the data quality of the geographic information is low.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart diagram of a method for recommending order of courier dispatch according to the present invention;

FIG. 2 shows correspondence between dispatch addresses, keyword vectors, latitude and longitude information, and points of interest in the process of converting relevant information of a dispatch process into points of interest in a specific embodiment;

FIG. 3 shows correspondence between input and output vectors as classifier training data in a specific embodiment;

fig. 4 shows a workflow of the dispatch sequence recommendation system based on courier history data according to the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the invention, fall within the scope of protection of the invention.

As shown in fig. 1, a flowchart of a method for recommending delivery order of couriers according to an embodiment of the present invention includes the following steps:

s1, acquiring historical dispatch data of an express delivery person;

The historical dispatch data for couriers available to a logistics company may include many aspects of data such as shipping numbers, dispatch addresses, dispatch dates, telephone numbers, etc. we can choose from the historical dispatch data associated with dispatch orders that may have an impact on dispatch orders.

For example, we select related information of several dispatch processes, each including at least a dispatch address and a dispatch order involved in performing several dispatch tasks. According to the correlation between the logistics information, the dispatch address and the dispatch sequence can also be obtained through other information association, for example, the corresponding dispatch address can be obtained through the bill number, the customer name and the customer telephone number.

It should be noted that, in the actual working process of the courier, one courier will be generally arranged to dispatch a plurality of couriers, and the process of dispatching each courier will be referred to as a dispatch process from the execution of the first dispatch task to the completion of the last dispatch task. For example, a dispatch process for a courier includes 6 dispatch tasks. The dispatch addresses related to the 6 dispatch tasks are arranged in the dispatch order as shown in table 1.

TABLE 1 delivery addresses and delivery orders involved in a delivery process for an courier

The step S2 of converting the historical dispatch sequence data of the courier into a training set of the model further includes the following steps:

s2.2, calling an interest point judgment model to obtain an interest point;

The interest point judgment model is well documented in the prior art, is not an invention point of the invention, and can be selected to obtain the interest point. Here we choose a specific point of interest decision model, the principle of which is roughly: the interest point is a set of a series of addresses, and when a new address is input, it is determined from which set the address is closer, and the distance is measured based on the navigation distance and the similarity of the text of the address.

The input vector of the interest point judgment model is a joint vector formed by converting a dispatch address into a keyword vector and longitude and latitude information, wherein the conversion from the dispatch address to the word vector is completed by means of an address word stock, and the conversion from the dispatch address to the longitude and latitude information is completed by means of an electronic map interface. The address word library stores all levels of address information words, and after the address is input into the address word library, keywords in the dispatch address are extracted, so that corresponding keyword vectors are found. The address-to-longitude-latitude information may use an interface provided by an electronic map service provider, such as a Goldmap, a hundred-degree map. The reason for using the combined vector composed of the keyword vector and the longitude and latitude information is that the keyword vector and the longitude and latitude information can mutually prove and complement each other, and because the address accuracy of the interest point is high, the accurate interest point can not be obtained by means of the longitude and latitude information obtained by the GPS, and therefore the keyword information is required to be used as the supplement to improve the accuracy of the interest point acquisition.

Taking the dispatch process of table 1 as an example, the 6 dispatch addresses in table 1 are firstly converted into keyword vectors and longitude and latitude information, and a joint vector formed by the keyword vectors and the longitude and latitude information is input into an interest point judgment model to obtain interest points corresponding to the 6 addresses respectively. The correspondence of the dispatch address, the keyword vector, the latitude and longitude information and the interest point is shown in fig. 2. For example, the first address in table 1 is converted into a joint vector consisting of a keyword vector and latitude and longitude information, which is (0, 1, 0) (114.7466, 22.34538). Finally, the obtained interest points are arranged according to the sending sequence, and then the interest point sequence list is obtained. Taking the dispatch process of table 1 as an example, the final interest point sequence list is POI1- > POI2- > POI3- > POI4.

We need to extract the input vector and the output vector from the point of interest sequential list. In step S2.4, the input vector is a joint vector formed by the quantity of the respective interest points and the current position of the courier before the courier performs one of the specific dispatch tasks in a specific dispatch process, and the output vector is the interest point to which the courier actually performs the specific dispatch task. The specific conversion principle is illustrated by taking the dispatch process of table 1 as an example: when the courier performs the first dispatch task, the quantity of the pieces of each interest point on the courier is (2,2,1,1), the current position is the starting position, the points do not belong to any interest points, the artificial definition is (0, 0), the joint vector formed by the quantity of the pieces of each interest point and the current position of the courier is (2,2,1,1) (0, 0), the interest point to which the courier performs the first dispatch task is POI1, and the output vector is POI1.

When the courier completes the first dispatch task, that is, before the second dispatch task is executed, the quantity of the parts of each interest point on the courier is (1,2,1,1), the current position of the courier is at the first address in table 1, that is, belongs to POI1, the current position vector is (1, 0), the number in the current position vector represents whether the number belongs to the corresponding interest point, the joint vector formed by the quantity of the parts of each interest point and the current position of the courier is (1,2,1,1) (1, 0), the interest point to which the courier executes the second dispatch task is to be POI1, and the output vector is POI1.

Before executing the third dispatch task, the amount of each interest point on the person is (0,2,1,1), the current position of the courier is at the second address in table 1, and belongs to POI1, the current position vector is (1, 0), the joint vector formed by the amount of each interest point and the current position of the courier is (0,2,1,1) (1, 0), the interest point to which he executes the third dispatch task is to be sent is POI2, and the output vector is POI2.

From the above description, we extract 6 sets of input vectors and output vectors from the point of interest sequence list as training data, their correspondence is shown in fig. 3. The above description is only a method for converting the related information of one dispatch process into training data, and we can refer to the method to convert the related information of a plurality of dispatch processes to obtain a huge amount of training data, namely a training set.

In the following, we train a classifier by using a training set, the classifier uses a random forest, and the random forest is a machine learning model based on a decision tree and can be used for solving the problem of multiple classification. The random forest is input in such a way that before a specific dispatch process is executed by an express delivery person, a joint vector formed by the quantity of each interest point and the current position of the express delivery person is output as the interest point to which the express delivery person actually executes the specific dispatch task.

And the classifier is trained to obtain a prediction model. The predictive model may be used for new dispatch tasks to obtain a list of recommended dispatch orders.

Step S3 further comprises the steps of:

Wherein, the current state information at least comprises:

The input vector of the prediction model is a joint vector formed by the quantity of the parts of each interest point before the courier executes one of the dispatch tasks and the current position of the courier in the dispatch process.

For example, the courier needs to complete 6 dispatch tasks in one dispatch process, and he wants to get the recommended dispatch sequence for the dispatch process. The dispatch addresses for these 6 dispatch tasks are shown in table 2.

Table 2 dispatch addresses for all dispatch tasks

When the dispatch process starts, the current state information of the courier is as follows: the courier does not perform any dispatch tasks, and all dispatch addresses related to the rest of the couriers carried by the courier are addresses shown in table 2, and the current position of the courier is in the logistics company. We now need to translate his current state information into input vectors for the predictive model.

Firstly, we call the interest point judging model to obtain the interest point based on the dispatch address, and the result is POI1, POI2, POI3 and POI4, wherein POI1 appears 2 times, POI2 appears 2 times, POI3 and POI4 appear once respectively, so we can know that POI1 has 2 pieces, POI2 has 2 pieces, and POI3 and POI4 have 1 piece respectively. The volume of each point of interest is (2,2,1,1). The current position of the courier is in the logistics company, and the address of the logistics company is known not to belong to any interest point, so that the current position of the courier is (0, 0). The joint vector of the two vectors is (2,2,1,1,0,0,0,0), and the point of interest to which the first step is recommended is POI1 by calling the prediction model with the joint vector as an input vector.

Now, the courier hopes to get the interest point to which the second step goes, and supposedly the courier follows the recommendation and goes to the POI1, the task of dispatching the first address in the table 2 is completed, and the current state information of the courier is updated as follows: all the dispatch addresses related to the remaining couriers carried by the courier are those shown in table 3, and their current location belongs to POI1. The number of the interest points is (1,2,1,1), the current position of the courier is (1, 0), and the interest point to which the second step is recommended is POI1 by taking (1,2,1,1,1,0,0,0) as an input vector and calling a prediction model.

TABLE 3 dispatch addresses for remaining tasks after completion of a dispatch task

Dispatch address
	Guangdong Shenzhen Longqiang Kaolin Xiyi 16 # Huatai toy (Shenzhen Co., ltd.)
Shenzhen Guangdong province Shenzhen city, longgang Ling Huangjin Haoting 2 span 722
	One of the golden south roads, golden Haotong, of the Dragon-sentry district, the Shenzhen city, guangdong
Shenzhen, guangdong, luo and Guangdong, shenzhen, longshun, gangji, high-grade middle school, and Xiao shop collection
	Guangdong Shenzhen city Longpost district bucky street south Lingyang garden a 3a 13a

Similarly, the current state information is updated once every time a dispatch task is completed, the point of interest to which we recommend the third step is POI2, the point of interest to which we recommend the fourth step is POI2, the point of interest to which we pay attention to is POI3, the point of interest to which we pay attention to is POI4, and the point of interest to which we pay attention to is POI1- > POI2- > POI3- > POI4 is the dispatch sequence recommendation list. Of course, updating the current state information is not necessarily performed after each delivery task is completed, and the default courier always performs the update according to the recommended sequence, so that a complete delivery sequence list can be obtained completely when the delivery process is not started yet. For example, if it is recommended that the point of interest to which the first step is to be POI1 and we default that the courier is to be directed to POI1 to complete the first dispatch task, we can infer that the current state information before he performs the second dispatch task is: all the dispatch addresses related to the rest of the express items carried by the courier are the addresses shown in table 3, the current position of the courier belongs to POI1, the quantity of the items of interest points is (1,2,1,1), the current position of the courier is (1, 0), the (1,2,1,1,1,0,0,0) is taken as an input vector, and the prediction model is called, so that the POI1 which is the point of interest and is recommended to go to the second step is obtained. We default the courier to POI1 to complete the second dispatch task and then infer the current status information before he performs the third dispatch task. That is, we can obtain the interest point recommended in the first step through the current state information when the dispatch process starts, then infer what state information the courier will update from the interest point recommended in the first step, and repeat the process to obtain the complete dispatch sequence recommendation list when the dispatch process does not start.

More generally, assuming a number of points of interest of n, this process can be expressed as:

Algorithm

Initialize:

l _i ＝0,i∈{1,2,...,n}

S＝(c ₁ ,c ₂ ,...,c _n ,l ₁ ,l ₂ ,...,l _n )

Whilenot all c _i ＝0do:

f(S)＝p

c _i ＝c _i -l _i ,i∈{1,2,...,n}

S＝(c ₁ ,c ₂ ,...,c _n ,l ₁ ,l ₂ ,...,l _n )

wherein S represents current state information, and f represents a prediction model; c _i Representing the piece quantity of a specific interest point;

representing the quantity of the parts of each interest point in the initial state; p represents an index of a specific interest point output by the prediction model, for example, p=2 represents an interest point 2; l (L) _i Indicating whether the actual position of the courier belongs to the specific interest point, 1 indicates that the courier belongs to the specific interest point, and 0 indicates that the courier does not belong to the specific interest point.

In the above specific embodiment, the training set used for training classification is extracted from the related information of several dispatch processes, and the related information of each dispatch process only includes the dispatch address and the dispatch sequence involved in executing several dispatch tasks. In practice, the relevant information of each dispatch process is not limited to the dispatch address and the dispatch sequence, and many information can influence the decision of the dispatch sequence, for example, when the ageing requirements of the products are different, the courier often chooses to send the products first with high ageing requirements, at this time, the applicability of the prediction model is reduced, and in order to improve the model, the accumulated information of the interest points of the products with different ageing requirements needs to be added as training data to train the classifier from the new state, that is, the relevant information of each dispatch process can also include the ageing requirements of the interest points. For example, when the classifier is trained regardless of aging, our input vector is a joint vector (2,2,1,1) (0, 0) composed of the number of pieces of interest and the current position of the courier, the output vector is POI1, when there is a courier with a POI1 aging requirement high, when considering the aging requirement, the input vector will become (2,2,1,1) (1, 0) (0, 0), the courier will go to POI1 when considering the aging requirement, the output vector is POI1, (1, 0) is the input vector converted from the aging requirement. The retrained predictive model may be adapted to new business scenarios. Therefore, the classifier can be trained by extracting proper data from the history data of the couriers and converting the proper data into input vectors serving as training data according to the requirements.

Next, a delivery order recommendation system based on courier history data is described, the system comprising:

current state information acquisition unit: the method comprises the steps of obtaining current state information of an express delivery person;

an information conversion unit: the method comprises the steps of converting current state information of an courier into an input vector of a prediction model;

The method for obtaining the training set of the prediction model and the training method of the classifier are described in detail in the express delivery order recommending method section, and are not described in detail here.

Specifically, the delivery order recommending system based on the historical data of the courier at least comprises the following information: in a dispatch process, before the courier performs one of dispatch tasks, the actual position of the courier; and in the dispatch process, before the courier executes one of the dispatch tasks, all dispatch addresses related to the rest of the couriers carried by the courier. The current state information acquisition unit at least provides an input interface of the real position of the courier and an input interface of the dispatch address.

As shown in fig. 4, when the system is used, the real position of the courier and all dispatch addresses related to the rest of the couriers carried by the courier are input.

The information conversion unit comprises an interest point judgment model, the information conversion unit firstly converts an input dispatch address into a joint vector composed of a keyword vector and longitude and latitude information, the interest point judgment model is called to obtain a plurality of interest points corresponding to the dispatch address, the piece quantity of each interest point is obtained according to the occurrence times of the interest points, the information conversion unit judges which interest point the real position of the courier belongs to, and the current position of the courier is obtained, wherein the joint vector composed of the piece quantity of each interest point and the current position of the courier is the input vector of the prediction model.

After the input vector of the prediction model is obtained, the prediction model is called, and the recommended interest point can be obtained. And calling a prediction model once every time the current state information is input until the dispatching task in a dispatching process is completed, so that a plurality of recommended interest points, namely dispatching forward and backward recommendation lists, can be obtained.

Of course, the current status information of the courier may also include a lot of content, not limited to the real location of the courier and all the dispatch addresses related to the remaining couriers carried by the courier. The current state information acquisition unit may also provide input interfaces for other current states accordingly during the continuous training and upgrading of the predictive model. For example, as described before, we add the point of interest accumulated information of products with different time efficiency requirements as training data to retrain the classifier, then the current state information acquisition unit can provide an input interface of the time efficiency information, so that the influence of the time efficiency information is added in the decision logic of the dispatch sequence. That is, when we want the rules of the recommended dispatch order to change, we can implement new dispatch order recommendation logic by retraining the classifier.

The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.

Claims

1. The express delivery order recommending method is characterized by comprising the following steps of:

s1, acquiring historical dispatch data of an express delivery person;

s2, converting historical dispatch sequence data of the courier into a training set of a model, and training a classifier by using the training set to obtain a prediction model; the training set for converting the historical dispatch sequence data of the courier into the model further comprises:

s2.2, calling an interest point judgment model to obtain an interest point;

s2.5, repeating the steps S2.1-S2.4 until all the dispatch processes are converted into training data, namely the training set;

s3, aiming at a dispatch process hoped to obtain a recommendation dispatch sequence, converting relevant information of the dispatch process into an input vector of a prediction model, and calling the prediction model to obtain a dispatch sequence recommendation list;

2. The method according to claim 1, wherein the historical dispatch data in step S1 includes information about a plurality of dispatch processes, and the information about each dispatch process includes at least a dispatch address and a dispatch sequence related to performing a plurality of dispatch tasks.

3. The method according to claim 1, wherein in the step S2.1, the input vector of the interest point determination model is a joint vector formed by converting the dispatch address into a keyword vector and longitude and latitude information.

4. The method according to claim 1, wherein in the step S2.4, the input vector is a joint vector formed by the amount of the respective points of interest and the current position of the courier before the courier performs one of the specific dispatch tasks in a specific dispatch process, and the output vector is the point of interest to which the courier actually performs the specific dispatch task.

5. The courier dispatch sequence recommendation method of claim 1, wherein the current status information comprises at least:

6. The method for recommending order for delivery of couriers according to claim 5, wherein the input vector of the prediction model is a joint vector composed of the quantity of the pieces of interest before the courier performs one of the delivery tasks and the current position of the courier in one of the delivery processes.

7. The courier dispatch sequence recommendation method of claim 1, wherein the classifier is a random forest classifier.

8. A delivery order recommendation system based on courier history data, comprising:

an information conversion unit: the input vector is used for converting the related information of the dispatch process into a prediction model; the information conversion unit is used for converting an input dispatch address into an input vector of the interest point judgment model, then calling the interest point judgment model to acquire interest points, arranging the acquired interest points according to a dispatch sequence involved in the dispatch process to obtain an interest point sequence list, and extracting the input vector from the interest point sequence list to serve as the input vector of the prediction model;

prediction model: the method comprises the steps of obtaining recommended interest points according to input vectors of a prediction model; the method comprises the steps of obtaining interest points to which a first step is recommended according to input vectors of a prediction model converted from current state information when a dispatch process starts; obtaining the interest point to which the next step is recommended according to the input vector of the prediction model converted from the updated current state information; arranging the interest points according to the acquisition sequence, and acquiring a sending sequence recommendation list;