Information Processing Apparatus, Information Processing Method, and Program

Abstract

To estimate a position bias in consideration of the influence of a bias even when the bias occurs in an item arrangement position.SOLUTION: An information processing device acquires an arrangement provability representing the provability that each of n items (n is a natural number of 2 or higher) is arranged in k positions (k is a natural number of 2 or higher), converts the n items to m embedded vectors (m is a natural number of 2 or higher) representing feature abstraction representations of the n items, calculates an allocation provability representing the provability of allocation to the m embedded vectors from the n items, and derives a provability representation where each of the m embedded vectors is arranged in each of the k positions using the arrangement probability distribution and the allocation provability distribution.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] In recent years, in web services used by users, rankings of all items such as products and search results are widely provided. In order to construct such rankings, click data such as click history by users is used. Since click data is not an explicit statement or evaluation by a user but is generated by the user's actions, it can be called implicit feedback. That is, click data can reflect how a user actually behaves when interacting with a system or website, rather than expressing likes and dislikes. From this point of view, click data is utilized to improve personalized rankings because it provides rich feedback implicitly.

[0003] When a user selects and clicks on an arbitrary item from a plurality of items displayed on a screen, since the user checks and clicks on the item, the position of the item can affect the check and click. Such a bias in the user's check according to the position of the item is called position bias. Non-Patent Document 1 discloses a technique for estimating position bias according to a regression-type EM (Expectation-Maximization) algorithm based on the position where an item is arranged and the click history of the item.

Prior Art Documents

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] On a screen where a plurality of items are displayed, each of the plurality of items is often arranged at a fixed position. For example, in carousel ads, the arrangement and order of each item are predetermined by the creator of the ad. Thus, in a situation where each of the plurality of items is arranged at a fixed position, a bias will occur in the positions where the items are arranged. That is, a bias occurs in the actual arrangement positions of the items with respect to the plurality of positions where the items can be arranged. The above document did not disclose a mechanism for estimating position bias while considering the influence of the bias in the arrangement position of items in such a case where a bias occurs in the arrangement position of items.

[0006] In view of the above problems, an object of the present disclosure is to establish an algorithm for estimating position bias while considering the influence of the bias even when a bias occurs in the arrangement position of items.

Means for Solving the Problems

[0007] An information processing apparatus according to an aspect of the present disclosure includes an acquisition unit that acquires an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions, a conversion unit that converts the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the characteristics of the n items, a calculation unit that calculates an assignment probability representing the probability of assignment from the n items to the m embedding vectors, and a derivation unit that derives a probability expression indicating that each of the m embedding vectors is arranged at each of the k positions using the distribution of the arrangement probability and the distribution of the assignment probability.

[0008] An information processing method according to an aspect of the present disclosure includes acquiring an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions, converting the n items into m (d is a natural number of 2 or more) embedding vectors representing an abstract expression of the characteristics of the n items, calculating an assignment probability representing the probability of assignment from the n items to the m embedding vectors, and deriving a probability expression indicating that each of the m embedding vectors is arranged at each of the k positions using the distribution of the arrangement probability and the distribution of the assignment probability.

[0009] A program according to an aspect of the present disclosure is a program for causing a computer to execute an information processing method, the information processing method including acquiring an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions, converting the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the characteristics of the n items, calculating an assignment probability representing the probability of assignment from the n items to the m embedding vectors, and deriving a probability expression indicating that each of the m embedding vectors is arranged at each of the k positions using the distribution of the arrangement probability and the distribution of the assignment probability.

Effect of the Invention

[0010] According to the present invention, an algorithm for estimating a position bias in consideration of the influence of the deviation of the arrangement position of items is provided.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2A

Figure 2B

Figure 2C

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiments for Carrying Out the Invention

[0012] Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. Among the components disclosed below, those having the same function are denoted by the same reference numerals, and the description thereof will be omitted. Note that the embodiments disclosed below are merely examples of means for realizing the present invention, and should be appropriately modified or changed according to the configuration of the apparatus to which the present invention is applied and various conditions, and the present invention is not limited to the following embodiments. Also, not all combinations of the features described in this embodiment are essential for the solution means of the present invention.

[0013] In this embodiment, a probability model obtained by improving the position-based click model that represents the probability model of the probability that a user clicks on an item, disclosed in Non-Patent Document 1, is used. First, the conventional position-based click model disclosed in the document will be described, and then a configuration for estimating position bias according to a regression-type EM algorithm based on the model obtained by improving the conventional position-based click model will be described. Note that in the present disclosure, the term "probability" may be understood as the ratio at which the target event occurs, rather than referring to an unknown possibility.

[0014] [Conventional Position-Based Click Model] The position-based click model disclosed in Non-Patent Document 1 will be described. Here, it is assumed that a user clicks on any one of one or more items (for example, advertisements or products) recommended and displayed (arranged) on a display screen in relation to a web service. Let item i be the item to be recommended. Also, let variable C be the reward variable. When clicks are the target of the reward, when variable C is "1", it indicates that the displayed item has been clicked, and when variable C is "0", it indicates that the displayed item has not been clicked. Also, let user u be a user having a user context including one or more user attributes (information regarding the user device or the user himself / herself) specific to user u. Also, let position k be the position where an item is displayed at a plurality of displayable (arrangeable) positions.

[0015] In the position-based click model, the click probability P(C = 1|i, u, k) conditional on item i, user u, and position k is expressed as in Equation (1) as the multiplication of two potential probabilities.

Equation

[0016] In the following description, the two probabilities on the right side of equation (1) are represented as the relevance μ(i, u) between the item and the user and the position bias θ k respectively. That is, μ(i, u) = P(R = 1|i, u) and θ k = P(E = 1|k). In Non-Patent Document 1, the relevance μ(i, u) and the position bias θ k are estimated using a regression-type EM algorithm. In the regression-type EM algorithm, the expectation (E: Expectation) step of calculating the expected value and the maximization (M: Maximization) step of maximizing the expected value are alternately repeated to perform maximum likelihood estimation of the probabilities included in the probability model (that is, the relevance μ(i, u) and the position bias θ k ).

[0017] Although there may be multiple positions where an item can be placed on the display screen, generally, due to the experience and expertise of advertisers and marketing personnel, items are often placed in fixed positions. That is, referring to equation (1), at the multiple positions that position k can take, the position where item i is placed is often not all of the multiple positions. In such a case where there is a bias in the actual item placement position with respect to the multiple positions where an item can be placed, the conventional position-based click model cannot accurately represent the position bias, and an accurate position bias cannot be estimated using the regression-type EM algorithm.

[0018] In this embodiment, a plurality of items are converted into a plurality of embedding vectors, and an improved position-based click model is defined based on the plurality of embedding vectors. Then, the relevance between the item and the user and the position bias represented by the improved position-based click model are estimated using the regression-type EM algorithm.

[0019] [Configuration of the Information Processing System] FIG. 1 shows a configuration example of the information processing system 1 in this embodiment. As an example, the information processing system 1 includes, as shown in FIG. 1, an information processing device 10 and a user device 11 used by a user. The information processing device 10 and the user device 11 are configured to be communicable. In the following description, although one user device 11 is shown in FIG. 1, a plurality of user devices may be configured to be communicable with the information processing device 10. Also, the term user device 11 may be understood to be any of the plurality of user devices.

[0020] The user device 11 is a device having a display unit, such as a smartphone, a tablet terminal, or a smart TV. The user device 11 is configured to be communicable with the information processing device 10 via a public network such as 5G (the fifth-generation mobile communication system) or a wireless communication network such as a wireless LAN (Local Area Network). When the user device 11 is a device having a GUI (Graphic User Interface) equipped on a display unit such as a liquid crystal display, like a smartphone or a tablet terminal, the user can perform various operations through the GUI. The operations include various operations on contents such as images displayed on the display screen, such as tap operations, slide operations, and scroll operations, using a finger, a stylus, or the like.

[0021] Note that the user device 11 is not limited to a device in the form shown in FIG. 1, and may be a device such as a desktop PC (Personal Computer) or a notebook PC. In that case, the operations by each user can be performed using an input device such as a mouse or a keyboard. Also, the user device 11 may separately include a display unit.

[0022] The information processing device 10 may be a server device that provides an electronic commerce platform such as a marketplace, and the user device 11 can utilize web services (Internet-related services) provided by the information processing device 10. Note that the information processing device 10 is not limited to being the above server device, and the user device 11 may be configured to utilize web services provided from a server device (not shown) different from the information processing device 10 via the information processing device 10.

[0023] The user can click (select) any of the multiple items (e.g., advertisements or products) provided by the web service and displayed on the display screen of the user device 11. If the clicked item is a product, after the click, product descriptions and information for the purchase procedure may be displayed on the display unit of the user device 11. Also, if the clicked item is an advertisement, the specific content of the advertisement may be displayed on the display unit of the user device 11.

[0024] The information processing device 10 is configured to provide a web service to the user device 11, observe the user's actions in the web service, and receive a report thereof. For example, the information processing device 10 can obtain observation data reflecting the user's action history in the web service by effectively setting the reception of a report indicating the user's action history. In this embodiment, the information processing device 10 observes the user's action history including the click operation by the user on any of the multiple items displayed on the display unit of the user device 11. The observation data obtained by the information processing device 10 includes at least any one of user context, information on the position of the item, item information of the item, and click information.

[0025] The user context includes one or more user attributes specific to the user (information regarding the user device 11 or the user himself / herself). The user attributes include, for example, the user's name, the user's address, information on the product delivery destination, information on the credit card held by the user, and the user's demographic information. The demographic information is information indicating demographic user attributes such as gender, age, residential area, occupation, and family composition. The user attributes can be registered by the user, for example, for using the web service. In addition to or instead of this, the information processing device 10 can obtain the user attributes by analyzing the web pages browsed by the user, the places clicked in the past, and the like.

[0026] The information on the position of the item includes information on the position of the item clicked by the user on the display unit of the user device 11. The item information includes information for identifying the item. Further, the item information may include one or more item features such as color and size. The click information is information indicating the presence or absence of a click.

[0027] Based on such observation data and predetermined configuration information regarding the display of items, the information processing device 10 generates a data set. The configuration information includes information on a plurality of items (including item features) displayed on the display screen of the user device and information on all positions (a plurality of positions) where the items can be arranged. Then, the information processing device 10 converts the plurality of items included in the data set into a plurality of embedding vectors and defines a position bias according to an improved position-based click model using the embedding vectors. The improved position-based click model corresponds to a model obtained by improving the conventional position-based click model of the above formula (1). Then, the information processing device 10 estimates the position bias using a regression-type EM algorithm. The regression-type EM algorithm in the present embodiment corresponds to an algorithm obtained by improving the regression-type EM algorithm disclosed in Non-Patent Document 1. In the following description, the former is also referred to as the improved regression-type EM algorithm, and the latter is also referred to as the conventional regression-type EM algorithm. Hereinafter, the improved position-based click model and the improved regression-type EM algorithm will be described in accordance with the operation of the information processing device 10.

[0028] [Operation of Information Processing Device] The information processing apparatus 10 acquires observation data over n times (n is a natural number of 2 or more) for the web service provided to the user apparatus 11, and generates a data set D based on the observation data and predetermined configuration information. One observation can be an observation over a certain period of time. In the present embodiment, it is assumed that the information processing apparatus 10 generates the data set D based on the observation data and the predetermined configuration information. However, the source data for generating the data set D is not limited to these, and other data may be used as long as the data set D can be generated by observing the click operation of the user.

[0029] The generated data set D is represented as in the following formula (2).

Equation

[0030] Let the set of actions including all possible (i, k) pairs be A = {(i, k)}, and let the function that maps user u to the distribution of the actions, in other words, the policy (rule) for placing item i at position k, be π. In this embodiment, the probability of placing (assigning) item i at position k (k ∈ K) (hereinafter also referred to as the placement probability) is represented by π(i, k). In many cases, the policy is often determined by a marketing person based on experience and expertise. As a result, the distribution of the placement probability π(i, k) is mostly deterministic and static. In this embodiment, the dataset D is a dataset that follows such a conventional policy, and it is assumed that the types of (i, k) pairs in the dataset D for all possible (i, k) pairs are limited.

[0031] When the information processing apparatus 10 acquires the dataset D, it calculates the placement probability π(i, k) for the dataset D. The placement probability π(i, k) for the dataset D represents the probability (ratio) at which item i included in the dataset D is placed at each position in the plurality of positions K where items can be placed. Also, as described above, in this embodiment, the types of (i, k) pairs in the dataset D are limited. The information processing apparatus 10 calculates an index indicating the bias of the distribution of the placement probability π(i, k) (hereinafter also referred to as the placement distribution index) in order to quantify the degree of such sparse pairs (i, k).

[0032] As a first example, when the placement distribution index is represented by the sparsity ratio J, the index can be expressed as in Equation (3). The sparsity ratio represents the ratio of non-missing values in the placement distribution (for example, the missing value refers to (i, k) missing in the dataset D).

Number

[0033] Also, as a second example, the placement distribution index can be represented by the similarity of the distribution of placement probabilities with respect to a uniform distribution in which n items i are uniformly placed for all possible positions k. An example of such similarity is the Kullback-Leibler divergence D KL and this index can be expressed as in equation (4).

Number

[0034] The sparsity rate J shown in equation (3) indicates the degree to which (i, k) is actually observed. Therefore, the smaller the sparsity rate J, the lower the proportion of (i, k) actually observed, indicating that the placement of items is sparse with respect to all possible placement positions. On the other hand, the larger the Kullback-Leibler divergence D KL shown in equation (4), the greater the deviation (dissimilarity) from the uniform distribution, indicating that there is a bias in the placement positions of the items. By calculating such a placement distribution index, the information processing apparatus 10 can grasp the degree of bias in the placement positions of the items in the dataset D.

[0035] Figure 2A shows an example of the distribution of the placement probability π(i,k) for the dataset D. Specifically, Figure 2A shows how an item is placed at each position with what probability (ratio) when all items are: item i0, i1, i2, and all positions where the items can be placed are: k0, k1, k2. The placement probability π b (i,k0), π b (i,k1), π b (i,k2), are shown as examples. In the table of Figure 2A, π biased (i,k) for the dataset D is represented as the placement probability π b (i,k). As shown in the table, in the dataset D, item i0 is fixedly placed at position k0, item i1 is fixedly placed at position k1, and item i2 is fixedly placed at position k2. That is, in the dataset D, it can be seen that the pairs of items and placement positions are biased towards (i0,k0), (i1,k1), (i2,k2).

[0036] For the distribution of the placement probability π(i,k) for the dataset D shown in Figure 2A, according to equation (3), the sparsity rate J for the dataset D can be calculated as in equation (5). [Number]

[0037] Also, according to equation (4), the Kullback-Leibler information quantity D KL for the dataset D can be calculated as in equation (6). [Number]

[0038] As described above, in the dataset D, for the (i, k) pairs, there is a bias for all possible (i, k). That is, in the (i, k) pairs in the dataset D, problems of bias and sparsity occur. And when using such a dataset D and equation (1) to estimate the position bias according to the prior art, due to the bias and sparsity problems of the (i, k) pairs included in the dataset D, there is a possibility that the estimation may not be performed correctly.

[0039] To address such problems, the information processing apparatus 10 generates a plurality of embedding vectors from a plurality of items in the dataset D. Specifically, the information processing apparatus 10 generates m embedding vectors e (e ∈ E, where E is a set of embedding vectors) from n items i j (j = 1 to n) in the dataset D. The embedding vector e corresponds to a vector representing latent contexts that abstract a plurality of item features.

[0040] The embedding vector e will be specifically described with reference to FIG. 3. FIG. 3 is a diagram for explaining the generation of the embedding vector according to the present embodiment. In the present embodiment, it is assumed that there are n items i, and each item is associated with a (a is a natural number of 2 or more) item features. The item features are features for identifying items such as color and size. First, the information processing apparatus 10 prepares an n × a matrix (item × item feature) from the n items i included in the dataset D. Each column of the n × a matrix corresponds to a feature vector, and each feature vector represents each item feature of each of the n items.

[0041] The information processing apparatus 10 converts (maps) an n×a matrix (item×item feature) into an n×m matrix (item×latent context). This conversion can be performed by known feature expression extraction techniques such as LSI (Latest semantic indexing) and VAE (Variational auto-encoder). Each column of the n×m matrix after conversion corresponds to an embedding vector, and each embedding vector represents the latent context of each of the n items. As a result, m embedding vectors are generated. The embedding vector corresponds to a vector obtained by abstracting a item features, and can also be referred to as an abstracted feature vector.

[0042] By converting the n×a matrix into an n×m matrix using LSI or VAE, the dimension is reduced, that is, m is smaller than a. For example, in LSI, the dimension can be reduced by grouping items with similar meanings in the item features of multiple items. Also, in VAE, the dimension can be reduced by compression processing. Thus, since the size of the n×m matrix becomes smaller than the size of the n×a matrix, sparse (i,k) pairs in the dataset D are converted into denser (e,k) pairs.

[0043] The probability of assignment from item i to embedding vector e (the probability of embedding vector e given item i) is represented by the assignment probability P(e|i). Fig. 2B shows an example of the distribution of the assignment probability P(e|i) for the dataset D. Specifically, Fig. 2B shows an example of the distribution of the assignment probability P(e|i) when the total number of items: item i0, i1, i2, and the number of embedding vectors m = 2. In this embodiment, for each item i, P(e|i) is calculated such that the sum of the assignment probabilities P(e|i) (the sum of the conditional probabilities of assigning item i to embedding vector e) becomes 1. That is, Σ e∈E(e|i) = 1. For example, in Figure 2B, the assignment probability P(e0|i0) from position i0 to embedding vector e0 is 1 / 2, and the assignment probability P(e1|i0) from position i0 to embedding vector e1 is 1 / 2. Therefore, P(e0|i0) + P(e1|i0) = 1. For positions i1 and i2, P(e0|i1) + P(e1|i1) = 1 / 3 + 2 / 3 = 1 and P(e0|i2) + P(e1|i2) = 1 / 4 + 3 / 4 = 1, respectively.

[0044] The relevance between the item and the user shown on the right side of Equation (1) can be expressed as in Equation (7) using the assignment probability P(e|i) and the embedding vector e. [Number] Based on the assignment probability P(e|i) and Equation (7), the information processing apparatus 10 can sample the reward w from the click C according to the conditional probability of the following Equation (8). Equation (8) represents the probability that the reward w takes the value 1 when the assignment probability P(e|i) is multiplied by the click probability P(C = 1|i, u, k) shown in Equation (1). [Number]

[0045] As described above, the information processing apparatus 10 generates the embedding vector e and samples the reward w, thereby generating a dataset D containing the embedding vector e and the reward w as shown in the following Equation (9) from the dataset D. e can be generated. [Number]

[0046] Also, the conventional position-based click model shown in Equation (1) can be expressed as an improved position-based click model with an embedding vector, as in Equation (10), using the embedding vector e. [Number]

[0047] In the following description, the two probabilities on the right side of equation (10) are respectively the relevance μ(e, u) between the embedding vector and the user, and the position bias θ ek and are represented as follows. That is, μ(e, u) = P(R = 1|e, u), and θ ek = P(E = 1|k). The position bias θ shown in equation (1) k is distinguished from the position bias θ shown in equation (10). ek The position bias θ shown in equation (10) represents the probability that the user u recognizes each of the k positions where the m embedding vectors are arranged.

[0048] As described above, in this embodiment, for each item i, P(e|i) is calculated so that the sum of the allocation probabilities P(e|i) is 1, and P(e|i) is defined as in the following equation (11).

Equation

[0049] When two items i and i' are similar, it is assumed that the corresponding embedding space representations e i,l and e i’,l are also similar, and the distributions of the allocation probabilities P(e|i) and P(e|i') are also similar. Based on such an assumption, even when observing a specific pair of (item, position) such as (i0, k0) and (i1, k1) in the dataset D, the allocation probabilities P(e0|i0), P(e1|i0), P(e0|i1), and P(e1|i1) can be obtained. The dataset D eBy using this, for each of the embedding vectors e0 and e1, data for both positions (i.e., in the case of the embedding vector e0, (e0, k0) and (e0, k1)) can be utilized, so that problems such as bias and sparsity as described above can be solved.

[0050] For each position k in the data set D, the information processing apparatus 10 derives an arrangement probability π(e, k) representing a probability expression in which the embedding vector e is arranged, using the distribution of the arrangement probability π(i, k) and the distribution of the allocation probability P(e|i). In the present embodiment, the arrangement probability π(e, k) is derived (calculated) using the sum of the multiplications of the allocation probability P(e|i) and the arrangement probability π(i, k). FIG. 2C shows an example of the arrangement probability π(e, k) in the case of the allocation probability P(e|i as shown in FIG. 2B. The arrangement probability π(e, k) is based on the allocation probability P(e|i shown in FIG. 2B and the arrangement probability π b (i, k) in the data set D e shows the arrangement probability π(e, k) for the data set D.

[0051] The arrangement probability π(e p , k q ) can be calculated as the sum of the multiplications of the allocation probability P(e e |i) and the arrangement probability π p (i, k b ) for all items i in the data set D. q ) For example, in FIG. 2C, the arrangement probability π(e0, k0) can be calculated as in the following equation (12).

Equation

[0052] As shown in FIG. 2C, it can be seen that the problem of data bias and sparsity is eliminated as compared with the arrangement probability π b (i, k) shown in FIG. 2A. Therefore, the data set D including the embedding vector e and the position ke By using this, it is expected to more accurately estimate the position bias.

[0053] Next, the information processing apparatus 10 estimates the position bias using the data set D e Conventionally, the position bias shown in the conventional position-based click model represented by Equation (1) was estimated using the regression-type EM algorithm described in Non-Patent Document 1. That is, by applying the regression-type RM algorithm and repeating the expectation step and the maximization step, the relevance μ(i, u) and the position bias θ k were optimized.

[0054] In the present embodiment, as an improved regression-type RM algorithm, first, in the expectation step, in the iteration of t + 1 for a certain time t, the distributions of the hidden variables E and R are estimated from θ ek (t) and μ (t) (e, u). θ ek (t) and μ (t) (e, u) are, respectively, at time t, the position bias θ ek described with reference to Equation (10) and the relevance μ(e, u) between the embedding vector and the user.

Equation

[0055] (13), for all data points in the data set D e , the probabilities P(E = 1|u, e, w, k) and P(R = 1|u, e, w, k) can be calculated. The probability P(E = 1|u, e, w, k) represents the probability (= 1) that there is a relevance between the user u and the embedding vector e when conditioned on the user u, the embedding vector e, the reward w, and the position k. Also, the probability P(E = 1|u, e, w, k) represents the probability that the position k is recognized by the user u when conditioned on the user u, the embedding vector e, the reward w, and the position k.

[0056] In the maximization step, using the probabilities from the expectation step, θ ek (t+1) and μ (t+1) (e,k) are calculated.

Equation

[0057] The improved regression-type EM algorithm according to this embodiment based on equations (13) and (14) is shown in Figure 4. Hereinafter, the processing in the algorithm will be described in order. Process 1 As input, receive a dataset D including a user (user context) u, an item i, a click c, and a position k, a position bias: θ ek , the relevance between the embedding vector and the user: μ(e,u), and the assignment probability P(e|i). θ ek can be a predetermined initial value. Also, μ(e,u) can be an empty regression model. Processes 2 to 4 For all users u, items i, clicks c, and positions k included in the dataset D, sample a reward w from the click c with the assignment probability P(e|i). Specifically, according to equation (8), sample a reward w ∈ {0, 1}, that is, a reward w that takes 0 or 1. Process 5 From the set of rewards w sampled in process 4 and the dataset D, prepare (generate) a dataset D e including a user (user context) u, an item i, a reward w, and a position k. Processes 6 to 14 Repeat processes 7 to 13 until the conditions of process 14 are satisfied (repetition from time t to time t+1). Process 7 Set the set S to an empty set. Processes 8 to 11 Dataset D e For all users u, embedding vectors e, clicks c, and positions k included in, sample the relevance r that takes r ∈ {0, 1}, that is, 0 or 1, from the probability P(R = 1|u, e, w, k) based on equation (13). Subsequently, generate the union set S of the user u, the embedding vector e, the relevance r, and the set S. Process 12 Update μ(e, u) according to GBDT (Gradient Boosted Decision Tree) with μ(e, u) and S as inputs. Since the relevance between items and users can be non-linear, the GBDT method is used here to learn μ(e, u). Process 13 (Update θ according to equation (14). ek Process 14 If the difference in the values of θ updated at time t and time t+1 ek is less than or equal to a predetermined value, it is determined that the convergence condition is satisfied and the process ends. The predetermined value is, for example, 10 -3 . Here, in addition to the values of θ updated at time t and time t+1 ek , if the difference in the values of μ(e, u) updated at time t and time t+1 is less than or equal to a predetermined value, it may be determined that the convergence condition is satisfied and the process ends. Process 15 Return θ k and μ(e, u).

[0058] Thus, in this embodiment, even for the dataset D in which the types of pairs (i, k) indicating the arrangement positions of items are limited, the item i is converted into the embedding vector e, and the dataset D e ​is generated. Then, the position bias θ is estimated according to the improved regression-type EM algorithm shown in FIG. 4. ek Thus, the position bias θ can be estimated with higher accuracy than when estimating the position bias θ according to the conventional regression-type EM algorithm. k ek

[0059] When the information processing apparatus 10 estimates the position bias θ according to the improved regression-type EM algorithm shown in FIG. 4, the arrangement positions of the items for each user may be adjusted based on the estimated position bias θ. Specifically, first, the information processing apparatus 10 estimates the relevance μ(i, u) = P(R = 1|i, u) between the item and the user in the position-based click model shown in Equation (1) using the conventional regression-type EM algorithm, and identifies one or more items highly relevant to the user u. In addition to this, or instead of this, the information processing apparatus 10 may identify one or more items in descending order of relevance to the user u. ek ek

[0060] Subsequently, the information processing apparatus 10 estimates the position bias θ = P(E = 1|k) in the improved position-based click model shown in Equation (10) using the improved regression-type EM algorithm shown in FIG. 4. Then, the information processing apparatus 10 identifies the positions where the user u is likely to recognize based on the estimated position bias θ, and arranges (matches) the one or more identified items at the identified positions. As a result, the probability that the user u clicks on the one or more items increases, that is, the CTR (click-through rate) improves, and the advertising effect can be enhanced. ek ek

[0061] [Functional Configuration of Information Processing Apparatus] ​​​​​​FIG. 5 shows a configuration example of the information processing apparatus 10 according to the present embodiment. The information processing apparatus 10 includes a first data set generation unit 101, a bias calculation unit 102, an embedding vector generation unit 103, an assignment probability calculation unit 104, a second data set generation unit 105, a probability expression derivation unit 106, a relevance estimation unit 107, a position bias estimation unit 108, a content creation unit 109, and a content providing unit 110.

[0062] The first data set generation unit 101 generates a data set D that reflects the behavior history of a user in a web service. For example, the first data set generation unit 101 generates the data set D based on the observed data of the user's behavior history in the web service and predetermined configuration information regarding the display of items. As described above, the data set D is configured to include a user u, an item i, a click c, and a position k in each of the first to nth observations. The item i is associated with a item features. Further, the first data set generation unit 101 calculates and obtains an arrangement probability π(i, k) for the data set D. An example of the distribution of the arrangement probability π(i, k) for the data set D is shown in FIG. 2A.

[0063] The bias calculation unit 102 calculates an arrangement distribution index indicating an index indicating the bias of the distribution of the arrangement probability π(i, k) calculated by the first data set generation unit 101. In the present embodiment, as described above, the sparsity rate J defined by Equation (3) and the Kullback-Leibler information amount D KL are calculated. The smaller the value of the sparsity rate J, the more biased the arrangement position of the item is. On the other hand, the Kullback-Leibler information amount D KL indicates that the larger the value, the more biased the arrangement position of the item is.

[0064] The embedding vector generation unit 103 converts the n items i included in the dataset D into m embedding vectors e representing the abstract expressions of the item features of the n items i. In the present embodiment, as described with reference to FIG. 3, the embedding vector generation unit 103 prepares an n×a matrix (item×item feature) from the n items i included in the dataset D. Then, the embedding vector generation unit 103 converts (maps) the matrix into an n×m (item×latent context) matrix. The embedding vector generation unit 103 generates each column of the converted n×m matrix as an embedding vector e.

[0065] The assignment probability calculation unit 104 calculates an assignment probability P(e|i) representing the probability of assignment from the n items i included in the dataset D to the m embedding vectors e (the probability of the embedding vector e given the item i). An example of the distribution of the assignment probability P(e|i) for the dataset D is shown in FIG. 2B. The assignment probability calculation unit 104 can calculate P(e|i) such that the sum of the assignment probabilities P(e|i) for each item i is 1.

[0066] The second dataset generation unit 105 generates the dataset D e Specifically, as described above, the second dataset generation unit 105 samples a reward w that takes 0 or 1 from the click c with the assignment probability P(e|i) calculated by the assignment probability calculation unit 104 according to equation (8). Subsequently, the second dataset generation unit 105 generates a dataset D e including the user u, the item i, the reward w, and the position k from the set of rewards w and the dataset D.

[0067] The probability expression derivation unit 106 derives the placement probability π(e,k) representing the probability expression that the embedding vector e is placed, for each position k in the dataset D, using the distribution of the placement probability π(i,k) and the distribution of the assignment probability P(e|i). The probability expression derivation unit 106 can derive (calculate) the placement probability π(e,k) using the sum of the products of the assignment probability P(e|i) and the placement probability π(i,k). An example of the distribution of the probability expression is shown in FIG. 2C.

[0068] The relevance estimation unit 107 estimates the relevance μ(i,u) between the item i and the user u in the dataset D. Specifically, the relevance estimation unit 107 estimates the relevance μ(i,u) shown in equation (1) using a conventional regression-type EM algorithm.

[0069] The position bias estimation unit 108 estimates the position bias θ e with respect to the dataset D. ek Specifically, the position bias estimation unit 108 estimates the position bias θ shown in equation (10) ek using the improved regression-type EM algorithm shown in FIG. 4. Alternatively, or in addition to this, the position bias estimation unit 108 may estimate the position bias θ with respect to the dataset D. Specifically, the position bias estimation unit 108 estimates the position bias θ shown in equation (1) k using a conventional regression-type EM algorithm. k

[0070] The position bias estimation unit 108 may switch to estimating the position bias using either a conventional regression-type EM algorithm or an improved regression-type EM algorithm based on the placement distribution index calculated by the bias calculation unit 102. For example, when using the sparsity rate J as the placement distribution index, the position bias estimation unit 108, when the sparsity rate J is less than or equal to a predetermined value (the number of missing values in the placement distribution is greater than or equal to a predetermined level), the position bias θ shown in equation (10) ekcan be estimated using the improved regression-type EM algorithm shown in FIG. 4. On the other hand, when the sparsity rate J is greater than the predetermined value (the number of missing values in the placement distribution is less than the predetermined level), the position bias estimation unit 108 calculates the position bias θ shown in Equation (1). k can be estimated using the conventional regression-type EM algorithm. Also, as another example, when using the Kullback-Leibler information amount D as the placement distribution index KL the position bias estimation unit 108, when the information amount D KL is greater than or equal to the predetermined value (the bias of the placement distribution is greater than or equal to the predetermined level), calculates the position bias θ shown in Equation (10). ek can be estimated using the improved regression-type EM algorithm shown in FIG. 4. On the other hand, when the information amount D KL is less than the predetermined value (the bias of the placement distribution is less than the predetermined level), the position bias estimation unit 108 calculates the position bias θ shown in Equation (1). k can be estimated using the conventional regression-type EM algorithm. In this way, when there is no bias in the placement position of the items, without generating the dataset D e the position bias θ k is estimated from the dataset D by a conventional method, thereby reducing the processing load.

[0071] The content creation unit 109 creates content to be provided to the user u based on the relevance μ(i, u) between the item i and the user u in the dataset D estimated by the relevance estimation unit 107 and the position bias θ k or the position bias θ ek estimated by the position bias estimation unit 108. When the content is advertisement content including a plurality of advertisements (i.e., items), the content creation unit 109 sorts the plurality of advertisements in descending order of relevance to the user u based on the relevance μ(i, u). Then, the content generation unit 109 assigns the sorted advertisements to each position based on the position bias θ k or the position bias θ ek to create advertisement content. The content providing unit 110 provides the content created by the content creating unit 109 to the user u. For example, the content providing unit 110 causes the created content to be displayed on the display unit of the user device used by the user u.

[0072] As described above, in the information processing apparatus 10 according to the present embodiment, first, the first dataset generation unit 101 acquires the placement probability π(i,k) representing the probability that each of a plurality of items is placed at a plurality of positions with respect to the dataset D. Then, the embedding vector generation unit 103 converts the plurality of items into a plurality of embedding vectors. And the assignment probability calculation unit 104 calculates the assignment probability P(e|i) representing the probability of assignment from the plurality of items to the plurality of embedding vectors. Subsequently, the probability expression derivation unit 106 derives the placement probability π(e,k) indicating the probability expression in which each of the plurality of embedding vectors is placed for each of the plurality of positions, using the distribution of the placement probability and the distribution of the assignment probability. The position bias estimation unit 108 estimates the position bias θ shown in equation (10) ek using the improved regression type EM algorithm shown in FIG. 4.

[0073] Furthermore, in the information processing apparatus 10, the content creating unit 109 creates content so as to place an item highly relevant to the user at a position where the user is highly likely to recognize, and the content providing unit 110 can provide the generated content to the user.

[0074] An example of content in which a plurality of items are placed at a plurality of positions is shown in FIG. 6. In FIG. 6, the upper part shows the advertisement content 60 in which the advertisement items 611, 612, and 613 are fixedly placed at the positions 601, 602, and 603, respectively. Here, assuming that among the relevance between the user u and each advertisement item estimated by the relevance estimation unit 107, the relevance between the user u and the advertisement item 612 is the highest, and the relevance between the user u and the advertisement items 613 and 611 follows in order. Furthermore, the position bias θ for each position estimated by the position bias estimation unit 108 ek(or position bias θ k ) Assume that among them, the position bias for position 601 is the highest, followed by the position biases for positions 602 and 603. In this case, as shown in the lower part of FIG. 6, the content creation unit 109 creates the advertisement content 61 in which the advertisement items 612, 613, and 611 are assigned to positions 601, 602, and 603, respectively. Then, the content providing unit 110 provides the created advertisement content 61 to the user u. For example, the content providing unit 110 controls to display the advertisement content 61 on the display unit of the user device 11 of the user u.

[0075] On the display unit of the user device 11 of the user u, in the advertisement content 61, the more interesting advertisement item 612 is arranged at the position 601 with a high position bias. As a result, not only does it become a more personalized display form for the user u, but also the probability that the user u clicks on the advertisement item 612 increases, the CVR (conversion rate) improves, and effective marketing can be realized.

[0076] [Hardware Configuration of Information Processing Apparatus 10] Next, an example of the hardware configuration of the information processing apparatus 10 will be described. Also, the user device 11 may have a similar hardware configuration. FIG. 7 is a block diagram showing an example of the hardware configuration of the information processing apparatus 10 according to the present embodiment. The information processing apparatus 10 according to the present embodiment can be implemented on a single or multiple computers, mobile devices, or any other processing platform. Referring to FIG. 7, an example in which the information processing apparatus 10 is implemented on a single computer is shown, but the information processing apparatus 10 according to the present embodiment may be implemented in a computer system including a plurality of computers. The plurality of computers may be connected to be communicable with each other via a wired or wireless network.

[0077] As shown in FIG. 7, the information processing apparatus 10 may include a CPU (Central Processing Unit) 701, a ROM (Read Only Memory) 702, a RAM (Random Access Memory) 703, an HDD (Hard Disk Drive) 704, an input unit 705, a display unit 706, a communication I / F 707, a GPU (Graphics Processing Unit) 708, and a system bus 709. The information processing apparatus 10 may also include an external memory. The CPU 701 comprehensively controls the operations in the information processing apparatus 10, and controls each component (702 to 708) via the system bus 709 which is a data transmission path.

[0078] The ROM 702 is a non-volatile memory that stores control programs and the like necessary for the CPU 701 to execute processing. Note that the program may be stored in a non-volatile memory such as the HDD 704 or an SSD (Solid State Drive), or an external memory such as a removable storage medium (not shown). The RAM 703 is a volatile memory and functions as the main memory, work area, etc. of the CPU 701. That is, when executing processing, the CPU 701 loads necessary programs and the like from the ROM 702 into the RAM 703, and realizes various functional operations by executing the programs and the like.

[0079] The HDD 704 stores various data and various information necessary when the CPU 701 performs processing using a program, for example. Also, the HDD 704 stores various data and various information obtained when the CPU 701 performs processing using a program or the like, for example. The input unit 705 is composed of a pointing device such as a keyboard or a mouse. The display unit 706 is composed of a monitor such as a liquid crystal display (LCD). The display unit 706 may function as a GUI (Graphical User Interface) when configured in combination with the input unit 705.

[0080] The communication I / F 707 is an interface that controls communication between the information processing apparatus 10 and an external device. The communication I / F 707 provides an interface with a network and executes communication with an external device via the network. Various types of data, various parameters, etc. are transmitted and received between the external device via the communication I / F 707. In the present embodiment, the communication I / F 707 may execute communication via a wired LAN (Local Area Network) conforming to a communication standard such as Ethernet (registered trademark) or a dedicated line. However, the network available in the present embodiment is not limited to this, and it may be configured by a wireless network. This wireless network includes a wireless PAN (Personal Area Network) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). In addition, it includes a wireless LAN (Local Area Network) such as Wi-Fi (Wireless Fidelity) (registered trademark) and a wireless MAN (Metropolitan Area Network) such as WiMAX (registered trademark). Furthermore, it includes a wireless WAN (Wide Area Network) such as 4G and 5G. Note that the network may connect each device so that communication is possible, and as long as communication is possible, the communication standard, scale, and configuration are not limited to the above. The GPU 708 is a processor specialized for image processing. The GPU 708 can perform predetermined processing in cooperation with the CPU 701.

[0081] At least some of the functions of each element of the information processing apparatus 10 shown in FIG. 5 can be realized by the CPU 701 executing a program. However, at least some of the functions of each element of the information processing apparatus 10 shown in FIG. 5 may operate as dedicated hardware. In this case, the dedicated hardware operates based on the control of the CPU 701.

[0082] [User attributes] As described above, in this embodiment, the user context has one or more user attributes associated with the user. Here, examples of user attributes will be mentioned. User attributes include the devices the user has (user devices) and factual characteristics (factual information) about the user. Factual characteristics can be characteristics (information) based on facts that are actually or objectively obtained from the user devices or the user. In addition, user attributes may include user attributes (estimated user attributes) estimated by applying factual characteristics to a trained machine learning model. The machine learning model is configured to output, for example, the probability (corresponding probability) that each of a plurality of user attributes corresponds to (fits) the target user, with the factual characteristics of the target user as input. The estimated user attributes can be determined from the corresponding probabilities.

[0083] [Item Features] As described above, in this embodiment, an item is associated with a plurality of item features. Here, examples of item features will be mentioned. The features of an item may include information for identifying the item (item ID), information for identifying the genre (upper classification) of the item, information for identifying the shop where the item is sold (shop ID), etc. Item features can also include transaction information (such as the number of transactions) between the item ID and the genre ID, and between the item ID and the shop ID, according to the transaction history.

[0084] Although specific embodiments have been described above, these embodiments are merely illustrative and are not intended to limit the scope of the present invention. The devices and methods described in this specification can be embodied in forms other than those described above. Also, without departing from the scope of the present invention, appropriate omissions, substitutions, and changes can be made to the above-described embodiments. Forms with such omissions, substitutions, and changes are included in the scope of what is described in the claims and their equivalents, and belong to the technical scope of the present invention.

[0085] The disclosure of this embodiment includes the following configurations. [1] An acquisition unit that acquires an arrangement probability representing the probability that n items (n is a natural number of 2 or more) are arranged at k positions (k is a natural number of 2 or more); A conversion unit that converts the n items into m embedding vectors (m is a natural number of 2 or more) representing an abstract representation of the characteristics of the n items; A calculation unit that calculates an assignment probability representing the probability of assignment from the n items to the m embedding vectors; A derivation unit that derives a probability expression indicating that each of the m embedding vectors is arranged at each of the k positions, using the distribution of the arrangement probability and the distribution of the assignment probability; An information processing apparatus having the above.

[0086] [2] The information processing apparatus according to [1], wherein in the assignment probability, for each of the n items, the sum of the conditional probabilities of assigning the item to the embedding vector is 1.

[0087] [3] Each of the n items is associated with a features (a is a natural number of 2 or more), The conversion unit converts the n items associated with the a features into the m embedding vectors, where m is smaller than a. The information processing apparatus according to [1].

[0088] [4] The information processing apparatus according to any one of [1] to [3], further comprising an estimation unit that estimates a position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged, based on the assignment probability.

[0089] [5] A first estimation unit that estimates a first position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged, based on the assignment probability; The information processing apparatus according to any one of [1] to [4], further comprising a second estimation unit that estimates a second position bias representing the probability that a user recognizes each of the k positions where the n items are arranged, based on the arrangement probability.

[0090] [6] It further has a bias calculation unit that calculates the bias of the distribution of the arrangement probability, When the bias of the distribution of the arrangement probability is equal to or higher than a predetermined level, the first estimation unit estimates the first position bias, When the bias of the distribution of the arrangement probability is less than the predetermined level, the second estimation unit estimates the second position bias. The information processing apparatus according to [5].

[0091] [7] The bias calculation unit calculates, as the bias of the distribution of the arrangement probability, the ratio of the n items arranged among the k positions in the distribution of the arrangement probability, for the information processing apparatus according to [6].

[0092] [8] The bias calculation unit calculates, as the bias of the distribution of the arrangement probability, the similarity of the distribution of the arrangement probability with respect to the uniform distribution of the n items at the k positions, for the information processing apparatus according to [6].

[0093] [9] The bias calculation unit calculates the similarity by the Kullback-Leibler information amount, for the information processing apparatus according to [8].

Explanation of Signs

[0094] 1: Information processing system, 10: Information processing apparatus 10, 11: User apparatus, 101: First dataset generation unit, 102: Bias calculation unit, 103: Embedding vector generation unit, 104: Allocation probability calculation unit, 105: Second dataset generation unit, 106: Probability expression derivation unit, 107: Relevance estimation unit, 108: Position bias estimation unit, 109: Content creation unit, 110: Content providing unit

Claims

1. An acquisition unit that acquires an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions; A conversion unit that converts the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the characteristics of the n items; A calculation unit that calculates an assignment probability representing the probability of assignment from the n items to the m embedding vectors; An estimation unit that estimates a position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged, based on the assignment probability; An information processing apparatus, characterized by comprising the above.

2. The information processing apparatus according to claim 1, further comprising a creation unit that creates advertisement content in which a plurality of advertisement items are arranged at a plurality of positions based on the position bias.

3. An acquisition unit that acquires an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions; A conversion unit that converts the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the characteristics of the n items; A calculation unit that calculates an assignment probability representing the probability of assignment from the n items to the m embedding vectors; A first estimation unit that estimates a first position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged, based on the assignment probability; A second estimation unit that estimates a second position bias representing the probability that a user recognizes each of the k positions where the n items are arranged, based on the arrangement probability; An information processing apparatus, characterized by comprising the above.

4. The information processing apparatus according to claim 3, further comprising a creation unit that creates advertisement content in which a plurality of advertisement items are arranged at a plurality of positions based on the first position bias or the second position bias.

5. In the assignment probability, for each of the n items, the sum of the conditional probabilities of assigning the item to the embedding vector is 1. The information processing apparatus according to claim 1 or 3, characterized by this.

6. Each of the n items is associated with a (a is a natural number of 2 or more) characteristics, The conversion unit converts the n items associated with the a features into the m embedding vectors, where m is smaller than a, and the information processing apparatus according to claim 1 or 3 is characterized in that.

7. further comprising a bias calculation unit that calculates a bias of the distribution of the arrangement probabilities, when the bias of the distribution of the arrangement probabilities is equal to or greater than a predetermined level, the first estimation unit estimates the first position bias, when the bias of the distribution of the arrangement probabilities is less than the predetermined level, the second estimation unit estimates the second position bias, and the information processing apparatus according to claim 3 is characterized in that.

8. The bias calculation unit calculates, as the bias of the distribution of the arrangement probabilities, a ratio of the n items arranged among the k positions in the distribution of the arrangement probabilities, and the information processing apparatus according to claim 7 is characterized in that.

9. The bias calculation unit calculates, as the bias of the distribution of the arrangement probabilities, a similarity of the distribution of the arrangement probabilities to a uniform distribution of the n items at the k positions, and the information processing apparatus according to claim 7 is characterized in that.

10. The bias calculation unit calculates the similarity by Kullback-Leibler information amount, and the information processing apparatus according to claim 9 is characterized in that.

11. An information processing method executed by an information processing apparatus, acquiring an arrangement probability representing a probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions, converting the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the features of the n items, calculating an assignment probability representing a probability of assignment from the n items to the m embedding vectors, estimating, based on the assignment probability, a position bias representing a probability that a user recognizes each of the k positions where the m embedding vectors are arranged, and the information processing method is characterized by including.

12. An information processing method executed by an information processing apparatus, acquiring an arrangement probability representing a probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions, converting the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract expression of the features of the n items, Calculating an assignment probability representing the probability of assignment from the n items to the m embedding vectors; Based on the assignment probability, estimating a first position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged; Based on the arrangement probability, estimating a second position bias representing the probability that a user recognizes each of the k positions where the n items are arranged; An information processing method characterized by including the above.

13. A program for causing a computer to execute an information processing method, the information processing method including: Obtaining an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions; Converting the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract representation of the features of the n items; Calculating an assignment probability representing the probability of assignment from the n items to the m embedding vectors; Based on the assignment probability, estimating a position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged; A program including the above.

14. A program for causing a computer to execute an information processing method, the information processing method including: Obtaining an arrangement probability representing the probability that each of n (n is a natural number of 2 or more) items is arranged at k (k is a natural number of 2 or more) positions; Converting the n items into m (m is a natural number of 2 or more) embedding vectors representing an abstract representation of the features of the n items; Calculating an assignment probability representing the probability of assignment from the n items to the m embedding vectors; Based on the assignment probability, estimating a first position bias representing the probability that a user recognizes each of the k positions where the m embedding vectors are arranged; Based on the arrangement probability, estimating a second position bias representing the probability that a user recognizes each of the k positions where the n items are arranged; A program including the above.

Images