JP2005108011A - Information processor for generating transaction information, method for generating transaction information, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information processor capable of quickly generating transaction information to be recommended to users from whom respective entries are provided by combining a plurality of entries. <P>SOLUTION: The information processor is provided with a storage means for storing each of entries by a format of a numerical vector having a code-added value consisting of a positive/negative code and a transaction amount as a component, a search means for calculating the evaluation costs of combinations of entries stored in the storage means on the basis of the values of numerical vectors and searching an optimum combination by using the evaluation costs, and a matching processing means for generating the transaction information by performing matching processing in which transaction items are made to correspond to each other between different entries in the optimum combination searched by the search means. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus for generating transaction information to be recommended to a user who has provided each entry by combining a plurality of entries representing transactions of products or services desired by the user, a transaction information generation method, and Regarding the program.

  For example, when exchanging inventory between a plurality of stores, a so-called search system that finds a search condition that matches the search conditions of each user (each store) from a list of registered inventory is often used (for example, See Patent Document 1 below). Also, in the travel plan creation system, when combining transportation means such as trains, airplanes, taxis, etc., accommodation facilities such as hotels, sightseeing places, etc., the plan is basically based on individual searches, incorporating the wishes of each user. (See, for example, Patent Document 2 below).

  There are many cases where a good deal (relationship) is established for the first time when a plurality of business partners and customers are involved with each other. For example, in the stock exchange described above, there are many cases where there are a plurality of stocks to be obtained and stocks to be provided at each store of the trader. In such a case, it is difficult to find a business partner for each stock, and if the business partners are different from each other, the transportation cost is also increased. Depending on the inventory, there may be combinations such as obtaining together or vice versa.

By consolidating such transactions desired by a large number of traders and customers, a “many-to-many” combination that is preferable to the one-to-one or one-to-many case can be established. Here, “many” means several rather than many. In addition, since a transaction is involved, it is preferable that there are few related businesses and customers.
JP 2002-366807 A JP 2001-282893 A

  The present invention collects entries representing transactions of products or services desired by a user, generates transaction information by obtaining the matching at high speed, and can promptly recommend the information to the user and transaction information generation method As well as providing a program.

  In the information processing apparatus according to one aspect of the present invention, one transaction item is represented by a positive / negative code indicating whether a transaction for providing a transaction object or a transaction for obtaining a transaction object is desired, a transaction item name, and a transaction quantity. In the information processing apparatus for generating transaction information to be recommended to a user who provided each entry by combining a plurality of entries each including a plurality of the transaction items, each of the plurality of entries is represented by the positive / negative sign and Storage means for storing a signed value consisting of the transaction quantity as a component in the form of a numerical vector, and calculating an evaluation cost based on the value of the numerical vector for a combination of a plurality of entries stored in the storage means, Search means for searching for an optimal combination using the evaluation cost, and an optimal set searched by the search means By performing matching processing for associating the transaction items to each other across the middle of the combined different entry comprising a matching processing means for generating the transaction information.

  According to the present invention, each of the plurality of entries is expressed in the form of a numerical vector having a signed value consisting of a positive / negative sign and a transaction quantity as a component, and the evaluation cost is calculated based on the value of the numerical vector. Processing can be speeded up.

  Therefore, an information processing apparatus and a transaction information generation method capable of aggregating entries representing transactions of products or services desired by the user, generating transaction information by quickly seeking the matching, and promptly recommending to the user, and Can provide a program.

  FIG. 1 is a block diagram showing a schematic configuration of an electronic transaction system to which an information processing apparatus for generating transaction information is applied according to an embodiment of the present invention. The information processing apparatus according to the embodiment of the present invention is realized as the group recommendation server 2, for example. The group recommendation server 2 comprises a general-purpose computer, preferably a server computer, and is connected to the network 1. The group recommendation server 2 stores a program for realizing a function for generating or updating transaction information for recommendation to a user of a personal computer (PC) 3 or a portable terminal 4 connected to the network 1. It is stored in the device 25. This program includes, for example, three modules: an entry reception module 20, a recommended group search module 21, and a display module 22. The program storage device 25 is composed of, for example, a magnetic disk device or an optical disk device. The above program is read into a random access memory (RAM) under the control of a CPU (not shown) and executed by the CPU, whereby the computer can function as the group recommendation server 2. The group recommendation server 2 is also installed with an operating system that manages various computer resources and provides a file system, a network communication function, a graphical user interface (GUI), and the like. Further, the entry storage device 23 and the recommended group storage device 25 are realized on a file system that controls the magnetic disk device, the optical disk device, and the like.

  The group recommendation server 2 receives a plurality of entries transmitted through the network 1 from the user's PC 3 or mobile terminal 4. The user here is assumed to be a trader or a customer. The received entry is received by the entry receiving module 20 and stored in the entry storage device 23.

  For these entries, the group recommendation server 2 provides high-speed transaction information that represents a transaction relationship (matching) between a group consisting of a combination of entries that satisfy each entry's hope (expected utility) and the elements within the group. And recommend this to registrants of each entry in the group. Here, “element” means individual transaction items such as products and services. Elements may have attributes such as type, type, color, date and time, but in the following description, elements having different attribute values are treated as different elements. Each element is given a negative sign to indicate the desire to obtain (purchase) the element, and conversely, a positive sign to indicate the desire to provide (sell) the element. Is attached. For example, when the element A is represented with a sign of -A, it means that there is an entity (such as a trader) who desires a transaction for obtaining the element A.

  An entry is represented by a set of one or more signed elements. For example, e1 = {+ A, -B, -B, + C} represents one entry. This entry e1 represents the transaction content “I want to get two elements B and provide element A and element C”. Two signed elements having the same element name and different signs satisfy each other's condition for transaction. This can be matched. For example, if -C exists in the other entry e2, + C of e1 and -C of e2 can be matched. An actual transaction is defined between two signable elements that can be matched, or the transaction relationship is called matching.

  The “expected utility” of an entry refers to the ratio (value) of the number of elements that the trade wants to close. The expected utility may be an independent value for each entry, or the default value of the expected utility may be used unless otherwise specified. It is also possible to specify no expected utility at all. In that case, the expected utility can be regarded as zero. In addition, the ratio of the number of elements that can be regarded as the establishment of a transaction in the group search described in detail later is called the “utility value” of the entry. For example, if the expected utility of the entry e1 is 0.7 and the transaction for this entry e1, for example {-B, -B, + C} is successful, the utility value of e1 is 3/4 = 0.75, and the expected Since the utility value is exceeded, the transaction is interpreted as meeting the expected utility of entry e1.

  The number of signed elements in an entry is called its “entry size”. The size of the entry e1 is 4. The number of entries in a group is called the “group size”. The information on each group obtained by the group recommendation server 2 is then recommended (notified) to the registrant of each entry in the group. In addition, since the smaller group size is preferable, it is assumed that a constraint condition such as “must be 10 or less” is given in advance to the group size. A group that satisfies this size constraint condition (hereinafter referred to as “size constraint condition”) and whose utility values of the included entries are all equal to or greater than the expected utility is referred to as a “recommended group”. Obtaining a recommended group means obtaining a business relationship (matching) within the group.

  Here, for example, the following four entries e1 = {+ A, -B, -B, + C}, e2 = {+ A, -C, -D}, e3 = {+ B, -C, + E} , e4 = {− A, + D} has already been received by the group recommendation server 2, and a new entry e5 = {− A, + B, −E} is received there.

In the same entry, it may be assumed that a pair of elements having the same element name but different signs, such as + A and -A, is redundant and does not exist. Assuming this, the entry can be expressed as a numeric vector with elements as indices. For example, the entries e1, e2, ..., e5 above are
e1 = (1, -2, 1, 0, 0),
e2 = (1, 0, -1, -1, 0),
e3 = (0, 1, -1, 0, 1),
e4 = (-1, 0, 0, 1, 0),
e5 = (-1, 1, 0, 0, -1)
Can be expressed as a numeric vector. The index of each entry corresponds to the element names A, B, C, D, E. This numerical vector is referred to as an “element vector”.

  For a group of one or more entries, the sum of the absolute values of the components of the sum of the element vectors of those entries is called the “cost” of the group. For example, in the above example, the sum of the element vectors of the group {e1, e2, e3} is (2, -1, -1, -1, 1). The cost of this group is 2 + 1 + 1 + 1 + 1 = 6. In the process of calculating the sum of element vectors, the sum may be canceled by the positive and negative values of the same index. This means that transactions (matching) are possible between corresponding elements in different entries.

  For example, when the sum of the element vectors of the entries e1 and e2 is calculated, the sum value at the third index is 0. This means that a transaction is possible between element + C of entry e1 and element -C of entry e2.

  Generally, the lower the cost, the greater the utility value of each entry included in it. In particular, in the case of a group whose cost is 0, the utility value of any entry in the group is 1.

  Hereinafter, as a method for obtaining a recommended group, a first method for obtaining a recommended group for newly accepted entries will be described in the first embodiment, and a division into recommended groups for all entries is always performed. The second method will be described in the second embodiment. For example, the following four entries e1 = {+ A, -B, -B, + C}, e2 = {+ A, -C, -D}, e3 = {+ B, -C, + E}, e4 = Assume that {-A, + D} has already been registered, and entry e5 = {-A, + B, -E} is newly registered there. At this time, in the first method, as shown in FIG. 2, for example, G1 = {e1, e3, e5} is obtained as a group including the base point entry e5. The group recommendation server 2 recommends this group G1 to each registrant of e1, e3, and e5. Note that the utility value of each entry in group G1 is all 1. If the expected utilities of e2 and e5 are both 1/3 or less, the group {e2, e5} can also be recommended. On the other hand, in the second method, as shown in FIG. 3, for example, G1 = {e1, e3, e5}, G2 = {e2, e4} is obtained and recommended to the registrant of each related entry. Note that the second method may not be executed every time an entry is newly registered, but may be executed at certain intervals.

(First embodiment)
As described above, the first method is to obtain a recommended group for newly received entries in a situation where several entries have already been received and stored (registered) in the entry storage device 23. It is. This newly accepted entry as a reference for obtaining a recommended group is referred to as a “base entry”. In the first method, one or more recommended groups including a base point entry are obtained. Specifically, one or a plurality of groups including the base point entry and satisfying the group size constraint condition with the lowest possible cost are obtained.

  However, even for a group with the lowest cost, the utility value of each entry included in the group does not necessarily exceed the expected utility. Also, the expected utility is not always the same for every entry. Therefore, there are cases where it is not sufficient to minimize the cost. However, as will be described later, when determining one or a plurality of groups with the smallest possible cost and determining a specific matching for each of those groups, the difference between the expected utility and the actual utility value of each entry, By reducing in order from the largest, it is possible to control the recommended group to be easily obtained.

  Incidentally, the set of element names of interest is assumed to be specified in advance and that what corresponds to the index also advance any element name of the element vector is determined. It is assumed that each existing entry has been converted to an element vector at the time of acceptance. In addition, a set of target element names does not necessarily have to be specified in advance. If not specified, each time an entry is newly accepted, the element name newly used in the entry is added to the set of element names, and the number of digits of the element vector is dynamically increased accordingly. It is possible to respond by.

  In the first method, a group including only base point entries is set as an initial solution. The existing entries are added to, deleted from, or replaced with the initial solution, and the improvement is repeated sequentially. As one method of the successive improvement, for example, a simulated annealing method (SA method) is used in the present embodiment. Although the embodiment of the present invention can be realized using the basic function of the SA method, it is preferable to improve the performance by additionally using another function of the SA method.

In addition to the SA method, tabu search and genetic algorithms can also be used. These search methods are called meta-solutions (see, for example, Yanagiura and Ibaraki, “Combinatorial Optimization”, Asakura Shoten, 2001). As temperature parameters of the SA method, an initial temperature, a final temperature, the number of repetitions at the same temperature, and a coefficient to be applied when the temperature is lowered are given in advance. In the SA method, an operation of making a slight local change to the current solution candidate is called “move”.

  FIG. 4 is a flowchart showing a specific processing procedure for obtaining a recommended group according to the first method according to the first embodiment. This process consists of six steps as shown in FIG. Details of each step are as follows.

  First, in step S1, a base point entry is read, expressed as an element vector, and stored in the entry storage device 23. Next, in step S2, a group G consisting only of base point entries is created, and this is set as the initial solution. This group G is stored in the storage device 24 as the best solution (best group). Next, in step S3, group G is sequentially improved by the SA method.

  Here, the detailed processing procedure of step S3 is demonstrated in detail with reference to the flowchart of FIG.

  First, in S301, an initial temperature is set as the temperature variable t of the SA method. Next, in S302, 0 is set to the counter cnt of the number of repetitions at the same temperature. Next, in S303, the type of move for the group G is determined, and the difference in cost when the move is executed for the group G is calculated. There are three types of moves:

Type 1: Randomly select one entry e1 from outside G and add e1 to G.
Type 2: Select one entry e2 at random from inside G and delete e2 from G.
Type 3: Randomly select entries e1 and e2 from G inside and outside, and swap them. That is, e1 is deleted from G and e2 is added to G instead.

  Specifically, the process for determining and executing the type of move follows the following procedure.

(B): If the size of group G is 1, select type 1. When the size of the group G is equal to a predetermined size constraint condition, either type 2 or type 3 is selected at random. When the size of the group G is other than these, any of types 1, 2, and 3 is selected at random.

(B): Calculate the difference in cost when the move of the type selected in (a) is executed for group G.

(C): Stores the numbers e1 and e2 of the entries that were moved in (a) and the cost difference when the move is executed (these are used in the next S304).

  Next, in S304 of FIG. 5, if the difference of the cost calculated | required by S303 is 0 or less, following (i) will be performed, and if larger than 0, (ii) will be performed. Then, 1 is added to the counter cnt.

  (I): For group G, the move selected in S303 is executed. If the cost of group G is less than the cost of the best group, replace the best group with group G. If the cost of group G becomes 0, the process proceeds to step S4.

  (Ii): For the group G, whether to execute the move selected in S303 is determined based on the probability p = exp (−1 * cost difference / current temperature). Specifically, a random number of 0 or more and 1 or less is generated. If the value is p or less, the move selected in S303 is executed, and if it is greater than p, nothing is done.

  Next, in S305 of FIG. 5, the counter cnt is checked. If the counter cnt is less than the predetermined number of repetitions (max_cnt), the process returns to S303.

  Next, in S306, the temperature is lowered by multiplying the temperature variable t by a predetermined coefficient α (0 <α <1). If the temperature does not reach the given final temperature, the process returns to S302 (S307). When the final temperature is reached, the best group at this point is adopted as a new recommended group candidate in S308.

  Returning to FIG. If the number of recommended group candidates obtained in step S3 is less than the predetermined number of groups, the process returns to step S2, and if reached, the process proceeds to step S4. When returning to step S2, the random number table used in the SA method is replaced.

  Next, in step S4, specific matching is obtained for each candidate of the recommended group obtained in step S3. The utility value of each entry in the group is obtained for the matching result. Then, a group in which the utility value of each entry is greater than the expected utility is determined as a recommended group. This series of processing is called matching processing.

  Here, the detailed processing procedure of step S4 of FIG. 4 is demonstrated with reference to the flowchart of FIG. Here, one of the recommended group candidates is a group G, and the processing for this group G will be described in detail.

  First, in S401, the entries in the group G are classified for each element depending on whether the element has a positive element or a negative element. The result obtained by this classification is referred to as “entry information for each element” of group G.

  For example, suppose group G consists of the following nine entries. Each entry is represented by an element vector, and its index corresponds to element names A, B, C, D, and E in order from the left. The base entry is assumed to be e1.

e18 = (1, 1, 0, 0, 1),
e79 = (-1, 0, -2, -2, 0),
e80 = (1, -2, 0, 1, 1),
e72 = (-1, 1, 0, 1, -1),
e76 = (0, 1, -1, 1, 0),
e1 = (-1, -1, 0, 0, 0),
e96 = (0, 0, 1, 1, 0),
e63 = (0, 0, 0, -2, 1),
e88 = (0, 0, -1, 0, -1)
If these entries are classified by element according to whether they have positive elements or negative elements, they are as follows.

  If the element vector component has a value of -2 or less or 2 or more, the corresponding entry number appears multiple times.

Next, in S402, referring to the entry information for each G element, a portion where matching is inevitably determined is first determined. For example, in the case of the above example, there are two entry numbers in the + A column and three entry numbers in the -A column, so it can be seen that a matching partner is always found in + A of e18 and e80. . It is as follows when the part to be decided is specified.

  The column in which “# (number)” is entered is unconfirmed, and the entry in parentheses indicates that the matching partner is unconfirmed. These lists in parentheses are called “indeterminate lists”. For the unconfirmed list, in the steps after S404, the number on the right side of # is selected and confirmed. Specifically, rearrangement is performed after S404, and finally, entries in the same column are regarded as matching partners. If the positive column or negative column of the same element is blank, it is considered that both the positive and negative columns of the element are fixed. If the number of positive and negative entries matches as in element B, it is considered that the matching partner has been determined at this point.

  Next, in S403, for each entry in the group G, the utility value is obtained using only the portion determined in S402, and the utility value is subtracted from the expected utility (value) to obtain a difference value (expected utility-utility). Value). At this time, an entry having a utility value of 1 is regarded as an entry for which matching has been confirmed. Then, indeterminate entries are stored in the heap. This heap has a descending tree structure with respect to the difference between expected utility and utility value. The “heap” is a rooted tree, and each of the trees is satisfied so that the condition that “the parent element is greater than or equal to the element of v for each node v other than the root” is satisfied. An element of the ordered set S is assigned to the node v. The heap itself is publicly known, and for example, it is detailed in a reference document ("Information structure (above)", written by Takao Asano, Nippon Critics, p. 45). Note that “descending order” in a descending tree structure means that the parent element is “greater than or equal to” the child element.

ヒープが空でない限り、次のＳ４０４を繰り返し実行する。 In the case of the above example, assuming that the expected utility of any entry is 0.5, the utility value and the difference value between the expected utility and the utility value are as follows.

Unless the heap is empty, the next step S404 is repeatedly executed.

  In S404, the entry is deleted from the root of the heap (this entry is represented by the symbol ent). Then, the same entry as ent is searched from the unconfirmed list. If the same entry as ent is found, the next step (i) is executed, and if not found, (ii) is executed.

  (I): Replace ent with the first unconfirmed entry in the list and remove ent from the unconfirmed list. In addition, the utility value of ent and the difference value between the expected utility and utility value are updated. If the utility value is less than 1, ent is returned to the heap. When the utility value is 1, ent is confirmed and is not returned to the heap.

  (Ii): ent is determined (not returned to the heap).

In the above example, ent is e79 having the maximum difference between the expected utility and the utility value. The unconfirmed list is

に変更される。e79の効用値と期待効用と効用値の差分値については、 It is. e79 can be found in the -A column. Since e79 is the first unconfirmed entry in the -A column, no replacement is required.

Changed to For the difference between the utility value of e79 and the expected utility and utility value,

It turns into. Since the utility value of e79 is less than 1, return it to the heap. If the value on the right side of # in the unconfirmed list column becomes 0, the field is regarded as confirmed (even if an unconfirmed entry remains, the field is not regarded as an unconfirmed list).

  Next, in S405, if the difference value between the expected utility and the utility value is 0 or less for any entry, the group G is determined as the recommended group. Then, for the matching relationship between the elements in the entry, the entry information for each element of the group G is referred to, and the entries in the same column in the positive / negative column of the same element are determined to be matching partners. If there is an entry with a positive difference between the expected utility value and the utility value, group G is invalid (that is, not determined as a recommended group).

In the case of the above example, the utility value of each entry and the difference value between the expected utility and the utility value are as follows.

Since the difference values between the expected utility and the utility value are all 0 or less, this group G is determined as a recommended group. The entry information for each element of group G is as follows.

  Returning to FIG. In step S5, it is checked whether or not each recommended group obtained in step S4 can be divided into smaller recommended groups. If there is a recommendation group that can be split, split it and replace it with a small recommendation group.

  In general, e1 and e2 are “adjacent” with respect to M when there is a matching relationship between the elements of entries e1 and e2. If there is a sequence e1, e2, ..., en consisting of two or more entries for entry e1, en and any two adjacent entries are adjacent with respect to M, e1 and en are related to M “Connected”. The recommendation group obtained in step S4 may be divided into a plurality of equivalence classes by this equivalence relationship called connectivity. In general, the smaller the recommended group size, the better.

  In the next step S6, the recommended group search module 21 outputs recommended group data. That is, for each recommended group, a set of entry numbers included in the group and a matching relationship between elements between the entries are output.

In the above example, one of the recommended groups is
e18 = {+ A: e79, + B: e80, + E: e72},
e79 = {-A: e18, -C:-, -C:-, -D: e80, -D: e72},
e80 = {+ A: e1, -B: e18, -B: e72, + D: e79, + E:-},
e72 = {-A:-, + B: e80, + D: e79, -E: e18},
e76 = {+ B: e1, -C:-, + D: e63},
e1 = {-A: e80, -B: e76},
e96 = {+ C: e88, + D: e63},
e63 = {-D: e76, -D: e96, + E: e88},
e88 = {-C: e96, -E: e63}
Is output. The parentheses indicate the signed element name of the entry and the matching partner entry number for each element. Symbols "-" are attached to elements that have no matching partner. For example, +18 of e18: e79 indicates that + A of e18 and −A of e79 are in a matching relationship. Note that the recommended group information output in this way is stored in the recommended group storage device 24. Further, the display module 22 generates display information of the recommended group as described above, and transmits the display information to the PC 3 or the mobile terminal 4 of the corresponding user.

(Second Embodiment)
Next, the second method will be described. As described above, the second method divides all entries into recommended groups. The purpose of all entries is to recommend groups to as many entries as possible. Note that since there may be entries that do not belong to any recommended group, the term “division” is used for convenience, although it is not “division” in a strict sense. In the second method, all entries are divided into a plurality of groups satisfying the group size constraint condition, and the total cost of each group is minimized. Not all of the groups thus obtained are recommended groups, but, as in the case of the first method, control is performed so as to be as recommended as possible when determining matching within each group. In general, groups can be recommended for a large number of entries.

  As with the first method, a set of element names of interest is assumed to be specified in advance and that what corresponds to the index also advance any element name of the element vector is determined. Each entry is represented by an element vector at the time of registration.

  In the second method, all entries are first divided into groups consisting of only one entry. A group set obtained by this division is set as an initial solution. Then, the group set is sequentially improved while moving or exchanging entries between groups of the initial solution. As one method of the successive improvement, the SA method can be used as in the first method. The total cost of each group is used as the cost of the group set.

  FIG. 7 is a flowchart showing a specific processing procedure for obtaining a recommended group according to the second method according to the second embodiment. This process consists of five steps as shown in FIG. Details of each step are as follows.

  First, in step S1, all entries are divided into groups consisting of only one entry, and a group set G obtained as a result is used as an initial solution. The group set G is stored as the best solution (best group set).

  Next, in step S2, the group set G is sequentially improved by the SA method. Here, the detailed processing procedure of step S2 of FIG. 7 is demonstrated with reference to the flowchart of FIG.

  First, in S201, an initial temperature is set in the temperature variable t of the SA method. Next, in S202, 0 is set to the counter cnt of the number of repetitions at the same temperature. Next, in S203, two different groups g1 and g2 are randomly selected from the group set G, and the larger one is set as g1. If the size is the same, the smaller group number is g1. Then, the type of move for g1 and g2 is determined, and the difference in cost when the move is executed for group set G is calculated. There are four types of moves:

Type 1: Exchange entries e1 and e2 selected at random from g1 and g2 respectively between g1 and g2.
Type 2: Move randomly selected entry e1 from g1 to g2.
Type 3: Move randomly selected entry e2 from g2 to g1.
Type 4: Divide g1 into two groups by bisecting g1 entry number columns at randomly chosen boundaries.
Here, the process of determining and executing the type of move specifically follows the following procedure.

(B): If g1 is 1 (and therefore g2 is 1), select type 3. If the size of g2 is equal to the predetermined size constraint (and thus the size of g1 is also equal to the predetermined size constraint), type 4 is selected. When the size of g1 is equal to the predetermined size constraint condition and the size of g1 is less than the predetermined size constraint condition, either one of type 1 or 2 is selected at random. In other cases, one of types 1, 2, and 3 is selected at random.

(B): Calculate the difference in cost when the move of the type selected in (a) is executed for the group set G.

(C): The group numbers g1 and g2, the entry numbers e1 and e2, the boundaries in the type 4, and the cost difference when the move is executed are stored (these are used in the next S204).

  In S204, if the cost difference obtained in S203 is 0 or less, the following (i) is executed, and if it is larger than 0, (ii) is executed. Then, 1 is added to the counter cnt.

(I): For the group set G, the move with S203 selected is executed. If the cost of the group set G is smaller than the cost of the best group set, the best group set is replaced with the group set G. If the cost of the group set G becomes 0, the process proceeds to step S3.

(Ii): For the group set G, whether or not to execute the move selected in S203 is determined based on the probability p = exp (-1 * cost difference / current temperature). That is, a random number between 0 and 1 is generated. If the value is p or less, the move selected in S203 is executed, and if it is greater than p, nothing is done.

  In step S205, the counter cnt is checked. If the counter cnt is less than the predetermined number of repetitions (max_cnt), the process returns to S203.

  Next, in S206, the temperature is lowered by multiplying the temperature variable t by a predetermined coefficient α (0 <α <1). If the temperature does not reach the given final temperature, the process returns to S202 (S207). When the final temperature is reached, the best group set at this time is adopted as a candidate for the recommended group set (S208). If it is desired to obtain a plurality of recommended group set candidates, the process returns to step S1, the random number table used in the SA method is replaced, and steps S1 and S2 are repeated as many times as necessary.

  Returning to FIG. Next, in step S3, matching is determined for each group of the best group set G obtained in step S2, and the utility value of each entry in the group is obtained. Then, the entire group in which the utility value of each entry is greater than or equal to the expected utility is determined as the entire recommended group. The processing for each group is the same as S401 to S405 (see FIG. 6) in step S4 of the first method.

  Next, in step S4, it is checked whether or not each recommended group obtained in step S3 can be divided into smaller recommended groups. If there is a recommendation group that can be split, split it and replace it with a small recommendation group.

  In step S5, the recommended group search module 21 outputs data of all recommended groups. That is, for each recommended group, a set of entry numbers included in the group and a matching relationship between elements between the entries are output.

For example, the following recommended group data is output. g1, g2, and g3 are numbers assigned to the recommended groups.

(Third embodiment)
The third embodiment of the present invention can be combined with each of the first and second embodiments described above. In the first and second embodiments, matching between elements having the same element name and different signs is assumed. However, matching may be possible even if the element names are different. For example, a transaction that provides “Nagano Prefecture Vegetables” or “Kawakami Village Lettuce” to a person who wants to obtain “Koshinetsu Local Vegetables” can actually occur. In other words,-(Koshinetsu local vegetables) and + (Nagano prefecture vegetables),-(Koshinetsu local vegetables) and + (Kawakami village lettuce) can be matched even though they have different element names. In addition, lettuce is a category of vegetables, but there are cases where it is possible to provide “Vegetables from the Koshinetsu region” to those who want to obtain “Kawakami Village Lettuce”. This is because the element “+” (a vegetable in the Koshinetsu region) may mean that any vegetable in the Koshinetsu region can be provided.

  In the third embodiment, it is assumed in advance that there is a combination that can be matched between different elements, and such a combination is stored in a table or the like in advance. Then, when calculating the cost of the group (step S3 of the first method, step S2 of the second method) and determining the matching (step S4 of the first method, step S3 of the second method) Browse the table and add as many matches as possible.

  The method for calculating the group cost in this embodiment will be described in detail. First, taking into account matching between the same element names, the sum of these element vectors is taken. When there is a negative component element that can be matched with the positive component element of the sum vector S, 1 is subtracted from the positive component, and 1 is added to the negative component. Such processing is performed for all positive components of the vector S. Finally, the sum of the absolute values of the components of the vector S is obtained and used as the cost of the group.

For example, a group G consisting of the following nine entries will be described as an example. These entries are represented by element vectors, and the indexes correspond to element names A, B, C, D, and E in order from the left. It is assumed that there are three matching combinations (-A, + C), (-C, + E), and (-A, + E) between different elements. These are assumed to be stored in the table M = {(− A, + C), (−C, + E), (−A, + E)}.
e18 = (1, 1, 0, 0, 1),
e79 = (-1, 0, -2, -2, 0),
e80 = (1, -2, 0, 1, 1),
e72 = (-1, 1, 0, 1, -1),
e76 = (0, 1, -1, 1, 0),
e1 = (-1, -1, 0, 0, 0),
e96 = (0, 0, 1, 1, 0),
e63 = (0, 0, 0, -2, 1),
e88 = (0, 0, -1, 0, -1)
When calculating the cost of this group G, first, the sum of these element vectors is taken to obtain the vector S = (− 1, 0, −3, 0, 1). Next, with respect to the element E of the fifth component 1 of the vector S, it is checked in the table M whether there is a negative component element that can be matched with it. In this case, there are matching combinations (-C, + E) and (-A, + E). Therefore, for example, when (−C, + E) is adopted, the vector S becomes S = (− 1, 0, −2, 0, 0). Since the vector S no longer has a positive component, the cost of the group G is calculated as 1 + 2 = 3 at this point.

  Next, how to determine the matching for the recommended group candidates will be described in detail. First, matching between the same element names is determined by the same processing as in step S4 of the first method described above. Next, entries whose utility value is less than 1 are stored in a descending heap with respect to (expected utility-utility value). Then, one entry is extracted from the root of this heap, and a matching partner is searched with reference to the unconfirmed list and “a table of combinations that can be matched between different elements”. At that time, if no matching partner is found, the entry is taken out from the root of the heap again. If a matching partner is found, it is returned to the heap only if its utility value is less than 1. Thereafter, the same processing is repeated until the heap becomes empty. If, on the way, there is an unconfirmed list only in the positive element column, or an unconfirmed list only in the negative element column, the process ends there.

For example, it is assumed that the above-described group G is a recommended group candidate. First, when matching between the same element names is determined by the same process as step S4 of the first method described above, as entry information for each element of the group G,

Is obtained. Here, the parenthesis represents the unconfirmed list at this point. The utility value of each entry and the difference value between expected utility and utility value are as follows.

となり、e79, e80の効用値と、期待効用と効用値の差分値は、 Next, entries e79, e80, e72, e76 having utility values less than 1 are stored in the heap in descending order with respect to the difference value between the expected utility and the utility value. The root of this heap is e79. If e79 is taken out and e79 is searched from the unconfirmed list, e79 is found in the -C column. Next, a combination (-C, + E) including -C is found in the table M (if it cannot be found, the process returns to selecting an entry from the root of the heap). Then, it is checked whether there is an unconfirmed list in the + E column. In this case, since there is an unconfirmed list in the + E column and there is e80, matching is determined between + E of e80 and -C of e79 (if there is no unconfirmed list in the + E column) Return to the process of reselecting the entry from the root of the heap). As a result, the unconfirmed list

The utility value of e79, e80 and the difference between expected utility and utility value are

It becomes. Since the utility value of e79 is still less than 1, e79 returns to the heap. Since the utility value of e80 is 1, e80 is deleted from the heap.

  Thereafter, the same processing is repeated until the heap becomes empty. In this example, since there are only unconfirmed lists in the -A column and the -C column, the processing ends at that point.

  According to the embodiment of the present invention described above, each entry is represented by an element vector, and a sequential search is performed while evaluating a group at a cost calculated from the sum of the element vectors. The utility value is calculated by determining the business relationship (matching) between them. Therefore, specific matching information in the group is not required for group evaluation, and a good group candidate can be calculated at high speed. In addition, even when there are combinations that can be traded (matched) between different elements, such combinations are stored in advance in a table or the like, and the table or the like is referred to when determining the cost calculation or matching of the group. Thus, a good group can be calculated at high speed by adding matching as much as possible. Further, the electronic trading system described in the embodiment may be realized as a system for determining receipt of necessary supplies and unnecessary inventory between offices in the company, for example.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

The block diagram which shows schematic structure of the electronic transaction system to which the information processing apparatus which produces | generates transaction information is applied according to embodiment of this invention. The figure for demonstrating the outline | summary of 1st Embodiment. The figure for demonstrating the outline | summary of 2nd Embodiment. The flowchart which showed the specific process sequence which concerns on 1st Embodiment and calculates | requires a recommendation group according to the 1st method. The flowchart which shows the detailed process sequence of step S3 of FIG. The flowchart which shows the detailed process sequence of step S4 of FIG. The flowchart which showed the specific process sequence which concerns on 2nd Embodiment and calculates | requires a recommendation group according to a 2nd method. The flowchart which shows the detailed process sequence of step S2 of FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Network, 2 ... Group recommendation server, 3 ... Personal computer (PC), 4 ... Portable terminal, 20 ... Entry reception module, 21 ... Recommended group search module, 22 ... Display module, 23 ... Entry storage device, 24 ... Recommendation Group storage device, 25... Program storage device

Claims (12)

One transaction item is represented by a positive / negative sign indicating whether a transaction that provides a transaction object or a transaction that obtains a transaction object, a transaction item name, and a transaction quantity, each including a plurality of the transaction items In the information processing device that generates transaction information that should be recommended to the user who provided each entry by combining the entries of
Storage means for storing each of the plurality of entries in the form of a numerical vector having a signed value consisting of the positive / negative sign and the transaction quantity as a component;
Search means for calculating an evaluation cost based on the value of the numerical vector for a combination of a plurality of entries stored in the storage means, and searching for an optimal combination using the evaluation cost;
Matching processing means for generating the transaction information by performing matching processing that associates the transaction items with each other between different entries in the optimal combination searched by the search means;
An information processing apparatus comprising: Means for generating display information of the transaction information generated by the matching processing means;
Means for transmitting the display information to a user who provided each entry;
The information processing apparatus according to claim 1, further comprising: The information processing apparatus according to claim 1, wherein the numerical vector has an index corresponding to the transaction item name. The information processing apparatus according to claim 1, wherein the search unit calculates a sum of the numerical vectors, and uses an absolute value sum of each component of the numerical vector sum as the evaluation cost. Each of the entries includes a pre-specified expected utility value,
The search means calculates a utility value of each of the plurality of entries, and sequentially improves the combination of the entries so that a difference between the expected utility value and the utility value becomes small. The information processing apparatus described in 1. The information processing apparatus according to claim 5, wherein the utility value includes a transaction establishment ratio in one entry. The information processing apparatus according to claim 1, wherein the search unit searches for the optimum combination according to a simulated annealing method. The information processing apparatus according to claim 1, wherein the search unit searches for the optimum combination based on one entry newly provided by the user. The information processing apparatus according to claim 1, wherein the search unit divides all entries stored in the storage unit and searches for the optimum combination for each of the divided entries. A table associating transaction items with different transaction item names;
Means for adding a combination of entries including transaction items that can be matched according to the table to the optimal combination;
The information processing apparatus according to claim 1, further comprising: One transaction item is represented by a positive / negative sign indicating whether a transaction that provides a transaction object or a transaction that obtains a transaction object, a transaction item name, and a transaction quantity, each including a plurality of the transaction items In the transaction information generation method for generating transaction information to be recommended to the user who provided each entry by combining the entries of
A storage step of storing each of the plurality of entries in a storage means in the form of a numerical vector having a signed value consisting of the positive / negative sign and the transaction quantity as a component;
A search step for calculating an evaluation cost based on the value of the numerical vector for a combination of a plurality of entries stored in the storage means, and searching for an optimal combination using the evaluation cost;
A matching process step of generating the transaction information by performing a matching process that associates the transaction items with each other between different entries in the optimal combination searched by the search step;
Transaction information generation method including One transaction item is represented by a positive / negative sign indicating whether a transaction that provides a transaction object or a transaction that obtains a transaction object, a transaction item name, and a transaction quantity, each including a plurality of the transaction items In a program that generates transaction information that should be recommended to the user who provided each entry by combining
A function of storing each of the plurality of entries in a storage means in the form of a numeric vector having a signed value composed of the positive / negative sign and the transaction quantity as a component;
A search function for calculating an evaluation cost based on a value of the numerical vector for a combination of a plurality of entries stored in the storage means, and searching for an optimal combination using the evaluation cost;
A matching processing function for generating the transaction information by performing a matching process for associating the transaction items with each other between different entries in the optimum combination searched by the search function;
A program for realizing the above with a computer.

Images