CN113869950A - Takeaway platform restaurant sequencing method based on sales volume, user preference and preferential discount - Google Patents

Abstract

The invention discloses a takeaway platform restaurant sequencing method based on sales volume, user preference and preferential discount, which comprehensively considers the factors of the order sales volume of restaurants in recent time periods of a takeaway platform, the personal preference of a user, the preferential discount of the restaurants and the like, and realizes sequencing recommendation of restaurants near the user based on technologies such as data statistics, Singular Value Decomposition (SVD) and the like. The method comprises the following specific steps: relevant information of the restaurant is supplemented first: firstly, marking the dish classification of all restaurants; and secondly, calculating the comprehensive preferential discount of all restaurants. Second, recent orders are analyzed: firstly, determining restaurant order sales volume in different time periods; secondly, calculating the preference of all the users according to the historical orders by using an SVD matrix decomposition algorithm; then, the restaurant nearby the user is ranked and displayed by comprehensively considering the conditions of the restaurant, the distance between the restaurant and the user, the time of the user browsing the platform, the time-share order sales volume of the restaurant, the dish series preference of the user, the preferential discount of the restaurant and other factors.

Description

Takeaway platform restaurant sequencing method based on sales volume, user preference and preferential discount

Technical Field

The invention relates to the technical field of network information, in particular to a takeaway platform restaurant sequencing method based on sales volume, user preference and preferential discount.

Background

The restaurant ranking effect on the take-away platform directly affects the user experience, and the sales volume of merchants and the platform income often have a plurality of factors to be considered when ranking and recommending restaurants. First, the head restaurant with the high recent sales is an important order source of the platform and is also a reaction to the recent ordering trend of the user group, so the sales factor needs to be analyzed heavily in the ranking. Secondly, the preferences of the user on different cuisine are different, and in order to improve the order conversion rate, the cuisine preference factor of the user needs to be considered when the restaurant is displayed; meanwhile, preferential discount strength provided by the restaurant is often different, and the discount strength on the current day has a direct influence on the sales volume, so that preferential discount factors need to be considered in the sorting; how to comprehensively consider the above three factors in the sorting is a problem to be solved by the invention.

Disclosure of Invention

In order to solve the above problems, the present invention provides a takeaway platform restaurant ranking method based on sales volume, user preferences and preferential discount, which comprehensively considers factors such as recent restaurant order sales volume in different time periods of the takeaway platform, user's personal preferences, preferential discount of the restaurant and the like in the ranking process.

In order to achieve the purpose, the invention adopts the following technical scheme:

a takeaway platform restaurant ranking method based on sales volume, user preference and preferential discount is characterized in that:

step 1, restaurant data processing

The restaurant data processing method mainly relates to two aspects: marking the restaurants according to the cuisine categories to which the restaurants belong; calculating the current preferential discount strength of the restaurant;

1.1 restaurant cuisine Mark

The dish type is an important characteristic of the restaurant and is an important factor to be considered when the user carries out restaurant sequencing recommendation, so that the dish type of the restaurant needs to be marked in advance; a restaurant having one or more cuisine tags based on the dishes it primarily sells;

1.2 restaurant discount strength calculation

The restaurant may provide a plurality of full-range advantages, wherein the maximum advantage is required to be calculated as the final advantage of the restaurant, and the discount strength of the advantage of a restaurant is greater, and the restaurant should be preferentially shown from the viewpoint of the advantage;

step 2 order data analysis

Reading historical order data of the platform, wherein the data comprises the following fields: order ID, order time, user ID, restaurant ID, and restaurant cuisine;

2.1 time-phased sales statistics

Counting the order sales volume of each restaurant in time intervals by taking the read order data as a unit of hour;

2.2 calculating user cuisine preferences

Generating a hidden vector for each user and cuisine category by using a matrix decomposition algorithm, positioning the user and the cuisine on a space represented by the hidden vector, wherein the cuisine with a similar distance to the user shows that the user is more interested, so that a higher exposure rate is allocated to restaurants of the cuisine category; calculating preference values of the user for various cuisine in advance by using a Singular Value Decomposition (SVD) algorithm for subsequent sequencing;

step 3, restaurant screening and freight interval sorting

The restaurant screening and freight interval sequencing comprises the following two steps:

3.1 restaurant screening

The restaurant screening is divided into two steps, firstly, the real-time or input position information of the user is obtained, and only restaurants in the city where the user is located are reserved according to the position information; then, the restaurants in the off-shelf state are filtered, and only the restaurants in the business on the day are reserved;

3.2 freight calculation and sorting based on freight Interval

Calculating freight rates of the restaurants and the users, which are screened out from the 3.1, according to freight rate rules of the platform according to the positions of the users;

setting a plurality of freight intervals from low to high, and then marking the restaurants according to the freight intervals in which the restaurants are located;

sorting the restaurants from low to high according to the marked freight intervals by using the weak sequence relation;

step 4 order fusion

4.1 sales percentile order calculation

The integral point time of the user browsing platform is approximated, the restaurants in each freight interval of 3.2 are sorted again according to the corresponding integral point sales data in 2.1 by using the weak sequence relation, and the obtained results are recorded as sales sorting; then, calculating the percentage ranking of sales volume of all restaurants in the freight interval;

4.2 preference percentile computation

Inquiring the preference values of the user for different cuisine calculated in the step 2.2, sequencing the restaurants in each freight interval of 3.2 by using a weak sequence relation, and recording the obtained result as preference sequencing; then, calculating the percentage of preference ranking of all restaurants in the freight interval;

4.3 discount percentile calculation

Sorting the restaurants in each freight interval of 3.2 according to the final preference of the restaurants obtained by 1.2 by using a weak sequence relation, and recording the obtained result as a preference discount sorting; then, calculating the percentage of discount ranking preferential to all restaurants in the freight interval;

4.4 comprehensive percentile order calculation and restaurant rearrangement output

For restaurants in 3.2 freight zone intervals, summing three percentile orders obtained in 4.1, 4.2 and 4.3 and arranging the three percentile orders in the intervals in an ascending order;

step 5 data update

Taking into account the influence of seasonal factors on the order, it needs to be updated every two weeks for the ranking in 2.1, 2.2.

The preference value of the user for each cuisine is calculated in advance by using a Singular Value Decomposition (SVD) algorithm, which is as follows:

processing the read order data to generate a co-occurrence matrix M of a user ID and a restaurant cuisine, wherein the row of M represents a user, the column of M represents a cuisine, and the order placing times of a user i in the restaurant with a cuisine label of j in the past two-week history order are counted as an element M_ijTaking the value of (A); assuming that the matrix M is an M × n matrix, it is decomposed into M ═ U∑V^TWherein, U is an m multiplied by m orthogonal matrix, V is an n multiplied by n orthogonal matrix, and sigma is an m multiplied by n diagonal matrix, elements positioned on the sigma diagonal are called singular values, the singular values reflect the implicit important information in the matrix to a certain extent, and the value size is positively correlated with the importance; according to the size of the singular value on the diagonal matrix sigma, taking the k singular values with the largest numerical value in the diagonal matrix sigma as implicit characteristics, deleting other dimensionalities of sigma and corresponding dimensionalities in U and V, namely U_m×m、V_n×nAre respectively converted into U_m×k、V_k×nWhereby the matrix M is decomposed into M ≈ M' ═ U_m×k∑_k×kV_k×n ^TThus, matrix decomposition with implicit quantity dimension k is completed;

the matrix M 'is recorded, and each row of elements in the matrix M' represents a preference value of a user for different cuisine.

Compared with the prior art, the invention has the following advantages:

1) the restaurant sorting method comprehensively considers factors such as restaurant order sales volume, personal preference of users, preferential discount of restaurants and the like to sort the restaurants of the takeout platform;

2) in consideration of different eating habits of the consumers in the morning, the evening and the evening, the method of time-sharing statistics is used when the sales volume of the restaurant order is processed, and the sequencing effect is improved;

3) the SVD algorithm in machine learning is used for calculating the preference of the user to all cuisine (not all restaurants), so that the matrix sparseness problem existing in the traditional recommendation algorithm is well solved;

4) the displayed restaurant ranking list can be dynamically updated, the ranking results seen by different users due to different preferences are obviously different, and the display is prevented from being uniform.

The relevant data of a take-out platform is used for testing, and the improvement of the click rate, the order conversion rate and the order quantity is facilitated by the method.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Detailed Description

The invention is described in further detail below with reference to the figures and specific embodiments.

As shown in fig. 1, the present invention provides a takeaway platform restaurant ranking method based on sales volume, user preferences and preferential discounts, which is characterized in that:

step 1, restaurant data processing

Marking all restaurants in a restaurant table of a database of a foreign sales platform mainly relates to two aspects: marking the restaurants according to the cuisine categories to which the restaurants belong; and calculating the current preferential discount strength of the restaurant.

1.1 restaurant cuisine Mark

The cuisine category is an important characteristic of the restaurant and an important factor to be considered when the user makes restaurant ranking recommendation, so that the cuisine category of the restaurant needs to be marked in advance. The following categories can generally be distinguished: sweet milk tea, Yuexing, Macao, Shanghai, Chuanxiang, spicy, Japanese and Korean cuisine, breakfast snack, hamburger fast food, western dinner and the like. The excessive types of the cuisine can cause the problem of sparseness in subsequent matrix calculation, and about 10 are suggested. A restaurant may have one or more tags based on the dishes it primarily sells.

1.2 restaurant discount strength calculation

A restaurant may offer a variety of full offers, with the largest offer being calculated as the final offer for the restaurant, e.g., a restaurant offers a full discount of 15 minus 5, and also offers a full discount of 7 points, then the full discount offer is represented by (15-5)/15 ≈ 0.67, and the full discount is represented by 7/10 ≈ 0.7, and since 0.67<0.7, the final offer discount strength for the restaurant is represented by 0.67. In addition, if there is a situation that multiple full offers can be superimposed, the accumulated offer should be calculated according to the specific actual situation as the final offer of the restaurant.

Step 2 order data analysis

And only reading platform order data of the last two weeks in consideration of seasons and time efficiency. The following fields need to be included in the data: order ID, order time, user ID, restaurant cuisine.

2.1 time-phased sales statistics

And counting the order sales volume of all restaurants in the time periods of 30 minutes before and 30 minutes after each hour in the past two weeks by taking the read order data as the unit of hour. For example, the order sales at 12 noon calculates the respective cumulative sales for all restaurants in the 11:31 to 12:30 time period each day for the last two weeks. This data needs to be computed well in advance for use in sorting.

2.2 calculating user cuisine preferences

The matrix decomposition algorithm is a very classical sorting recommendation algorithm, a hidden vector is generated for each user and cuisine category, the user and the cuisine are positioned on a space represented by the hidden vector, cuisine with a close distance to the user shows that the user is more interested, and then restaurants of the cuisine can be given a higher exposure in the sorting process. Singular Value Decomposition (SVD) is a common method in matrix decomposition. The invention uses SVD algorithm to calculate the preference of the user for the dish series in advance.

Processing the read order data to generate a co-occurrence matrix M of a user ID and a restaurant cuisine, wherein the row of M represents a user, the column of M represents a cuisine, and the order placing times of a user i in the restaurant with a cuisine label of j in the past two-week history order are counted as an element M_ijThe value of (a). Assuming that the matrix M is an mxn matrix, it is decomposed into M ═ U ∑ V^TWherein, U is an m × m orthogonal matrix, V is an n × n orthogonal matrix, and Σ is an m × n diagonal matrix, elements on the Σ diagonal are called singular values, the singular values reflect important information implicit in the matrix to some extent, and the value size and the importance are positively correlated. According to the size of the singular value on the diagonal matrix sigma, taking the k singular values with the largest numerical value in the diagonal matrix sigma as implicit characteristics, deleting other dimensionalities of sigma and corresponding dimensionalities in U and V, namely U_m×m、V_n×nAre respectively converted into U_m×k、V_k×nWhereby the matrix M is decomposed into M ≈ M' ═ U_m×k∑_k×kV_k×n ^TThus, matrix decomposition with implicit quantity dimension k is completed;

the matrix M 'is recorded, and each row of elements in the matrix M' represents a preference value of a user for different cuisine.

Step 3, restaurant screening and freight interval sorting

When the user uses the sorting function, the restaurant screening is firstly executed, then sorting is carried out according to the freight rate interval, and the specific operation is as follows:

3.1 restaurant screening

The restaurant screening is divided into two steps, firstly, the real-time or input position information of the user is obtained, and only restaurants in the city where the user is located are reserved according to the position information. Then, the restaurants in the states of business suspension, off-shelf and the like are filtered, and only the restaurants in the business day are reserved.

3.2 freight calculation and sorting based on freight Interval

And calculating freight rates of the restaurants and the user generated by the restaurant and the user according to the freight rate rule of the platform for the restaurants screened in the 3.1 according to the position of the user.

Several freight intervals are set from low to high, such as interval a ═ a, B), interval B ═ B, C), interval C ═ C, d), and the like. The restaurant is then marked according to the freight interval in which it is located.

Weak ordering relation is a binary relation and has wide application in the ordering field. A binary relation is called a weak order relation if the following two conditions are met,

asymmetry:

negative transmission:

and (4) sorting the restaurants from low to high according to the marked freight intervals by using a weak ordering relation (dish ordering relation).

Step 4 order fusion

Based on 3.2 sorting based on freight intervals, the sales volume, preference and discount percentiles of different restaurants are calculated in each interval block.

Wherein, the percentage degree P (r) of a restaurant r in the freight interval is as follows:

wherein B (r) represents the number of restaurants ranked better than the restaurant r in the freight interval of the restaurant r, such as restaurants with sales higher than r, restaurants that the user prefers r in the preference of dishes, restaurants with discount strength higher than r, and the like. E (r) represents the number of restaurants with the same ordering as the restaurant, and N represents the total number of restaurants in the freight interval of the restaurant r.

4.1 sales percentile order calculation

Approximate the hour of the user browsing the platform, e.g., 11:31 to 12:30 using hour 12. And (3) sorting the restaurants in each freight interval of 3.2 according to the corresponding whole-point sales volume data in 2.1 by using the weak ordering relation (weak ordering relation) again, and recording the obtained result as sales volume sorting. The sales ranking can be viewed as a further rearrangement of restaurants in each freight interval, but with restaurants in the lower freight interval being kept entirely ahead of restaurants in the higher freight interval as a whole.

Using the above percentile calculation formula, the percentile ranking of sales of all restaurants in the freight interval in which the restaurants are located is calculated. For a restaurant r sales percentage, use P_S(r).

4.2 preference percentile computation

And (3) inquiring the preference values of the user for different cuisines calculated in the step 2.2, sequencing the restaurants in each freight interval of the step 3.2 by using a weak sequence relation, and recording the obtained result as preference sequencing. Likewise, preference ranking can be viewed as a further rearrangement of restaurants in each freight interval according to user preferences, but with restaurants in the lower freight interval being kept entirely ahead of restaurants in the higher freight interval as a whole.

Using the above percentile calculation formula, the percentile of the preference ranks of all restaurants in the freight interval in which the restaurants are located is calculated. For a restaurantPreference of r in percentile order, using P_P(r).

4.3 discount percentile calculation

And sorting the final preference of the restaurants in each freight interval of 3.2 according to 1.2 by using the weak sequence relation, and recording the obtained result as the preference discount sorting. Likewise, preferential discount ranking may be viewed as a further rearrangement by preferential discount of restaurants in each freight interval, but with restaurants in the lower freight interval being kept entirely ahead of restaurants in the higher freight interval as a whole.

And calculating the percentage of discount sequencing preferential discount in the freight interval of all the restaurants by using the above calculation formula of the percentage. For discount percentile of one restaurant r, use P_D(r).

4.4 comprehensive percentile order calculation and restaurant rearrangement output

For restaurants in 3.2 freight intervals, the three percentile values obtained in 4.1, 4.2 and 4.3 are summed and arranged in ascending order in the interval.

Specifically, the total percentile P for a restaurant r_G(r) calculated using the formula:

P_G(r)＝P_S(r)+P_P(r)+P_D(r)

and (4) for all the restaurants in each freight interval, rearranging and outputting the restaurants according to the comprehensive percentile ranking of all the restaurants. The comprehensive percentile rearrangement can be regarded as the rank adjustment of the restaurants in the freight interval, and the overall arrangement still keeps the restaurants in the interval with lower freight completely arranged in front of the restaurants in the interval with higher freight.

Step 5 data update

Taking into account the influence of seasonal factors on the order, it needs to be updated every two weeks for the ranking in 2.1, 2.2.

Claims (2)

1. A takeaway platform restaurant ranking method based on sales volume, user preference and preferential discount is characterized in that:

step 1, restaurant data processing

The restaurant data processing method mainly relates to two aspects: marking the restaurants according to the cuisine categories to which the restaurants belong; calculating the current preferential discount strength of the restaurant;

1.1 restaurant cuisine Mark

The dish type is an important characteristic of the restaurant and is an important factor to be considered when the user carries out restaurant sequencing recommendation, so that the dish type of the restaurant needs to be marked in advance; a restaurant having one or more cuisine tags based on the dishes it primarily sells;

1.2 restaurant discount strength calculation

The restaurant may provide a plurality of full-range advantages, wherein the maximum advantage is required to be calculated as the final advantage of the restaurant, and the discount strength of the advantage of a restaurant is greater, and the restaurant should be preferentially shown from the viewpoint of the advantage;

step 2 order data analysis

Reading historical order data of the platform, wherein the data comprises the following fields: order ID, order time, user ID, restaurant ID, and restaurant cuisine;

2.1 time-phased sales statistics

Counting the order sales volume of each restaurant in time intervals by taking the read order data as a unit of hour;

2.2 calculating user cuisine preferences

Generating a hidden vector for each user and cuisine category by using a matrix decomposition algorithm, positioning the user and the cuisine on a space represented by the hidden vector, wherein the cuisine with a similar distance to the user shows that the user is more interested, so that a higher exposure rate is allocated to restaurants of the cuisine category; calculating preference values of the user for various cuisine in advance by using a Singular Value Decomposition (SVD) algorithm for subsequent sequencing;

step 3, restaurant screening and freight interval sorting

The restaurant screening and freight interval sequencing comprises the following two steps:

3.1 restaurant screening

The restaurant screening is divided into two steps, firstly, the real-time or input position information of the user is obtained, and only restaurants in the city where the user is located are reserved according to the position information; then, the restaurants in the off-shelf state are filtered, and only the restaurants in the business on the day are reserved;

3.2 freight calculation and sorting based on freight Interval

Calculating freight rates of the restaurants and the users, which are screened out from the 3.1, according to freight rate rules of the platform according to the positions of the users;

setting a plurality of freight intervals from low to high, and then marking the restaurants according to the freight intervals in which the restaurants are located;

sorting the restaurants from low to high according to the marked freight intervals by using the weak sequence relation;

step 4 order fusion

4.1 sales percentile order calculation

The integral point time of the user browsing platform is approximated, the restaurants in each freight interval of 3.2 are sorted again according to the corresponding integral point sales data in 2.1 by using the weak sequence relation, and the obtained results are recorded as sales sorting; then, calculating the percentage ranking of sales volume of all restaurants in the freight interval;

4.2 preference percentile computation

Inquiring the preference values of the user for different cuisine calculated in the step 2.2, sequencing the restaurants in each freight interval of 3.2 by using a weak sequence relation, and recording the obtained result as preference sequencing; then, calculating the percentage of preference ranking of all restaurants in the freight interval;

4.3 discount percentile calculation

Sorting the restaurants in each freight interval of 3.2 according to the final preference of the restaurants obtained by 1.2 by using a weak sequence relation, and recording the obtained result as a preference discount sorting; then, calculating the percentage of discount ranking preferential to all restaurants in the freight interval;

4.4 comprehensive percentile order calculation and restaurant rearrangement output

For restaurants in 3.2 freight zone intervals, summing three percentile orders obtained in 4.1, 4.2 and 4.3 and arranging the three percentile orders in the intervals in an ascending order;

step 5 data update

Taking into account the influence of seasonal factors on the order, it needs to be updated every two weeks for the ranking in 2.1, 2.2.

2. The takeaway platform restaurant ranking method of claim 1 based on sales volume, user preferences, and preferential discounts, comprising: the preference value of the user for each cuisine is calculated in advance by using a Singular Value Decomposition (SVD) algorithm, which is as follows:

processing the read order data to generate a co-occurrence matrix M of a user ID and a restaurant cuisine, wherein the row of M represents a user, the column of M represents a cuisine, and the order placing times of a user i in the restaurant with a cuisine label of j in the past two-week history order are counted as an element M_ijTaking the value of (A); assuming that the matrix M is an mxn matrix, it is decomposed into M ═ U ∑ V^TWherein, U is an m multiplied by m orthogonal matrix, V is an n multiplied by n orthogonal matrix, and sigma is an m multiplied by n diagonal matrix, elements positioned on the sigma diagonal are called singular values, the singular values reflect the implicit important information in the matrix to a certain extent, and the value size is positively correlated with the importance; according to the size of the singular value on the diagonal matrix sigma, taking the k singular values with the largest numerical value in the diagonal matrix sigma as implicit characteristics, deleting other dimensionalities of the diagonal matrix sigma and corresponding dimensionalities in U and V, namely U_m×m、V_n×nAre respectively converted into U_m×k、V_k×nWhereby the matrix M is decomposed into M ≈ M' ═ U_m×k∑_k×kV_k×n ^TThus, matrix decomposition with implicit quantity dimension k is completed;

the matrix M 'is recorded, and each row of elements in the matrix M' represents a preference value of a user for different cuisine.

Images