JP2012238290A - Competing experience attributes indicating device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a user carrying out a marketing analysis to efficiently understand a competing subject by indicating features of a competing brand and competitive elements in terms of factors representing events having positively influenced the decision by a person to do what he or she experienced and factor attributes.SOLUTION: In order to appropriately identify "various viewpoints classified by a system of classification structured on the basis of the framework of marketing analysis with an appropriate number of viewpoints without overlapping" in classifying factor words by viewpoint in advance to analyze competing subjects according to different viewpoints in a form readily understandable by a user, factor attributes based on a definition of classification of consumer situations by Belk et al. are identified as a core of viewpoints, factor words are classified for the respective viewpoints, and competition analysis is performed for each viewpoint.

Description

  The present invention relates to a competitive experience attribute display apparatus, method, and program, and more particularly to a competitive experience attribute display apparatus that displays competitive experience attributes and related information using the similarity of factor topic distribution for each factor attribute based on a factor attribute table. And methods and programs.

  First, terms used in this specification will be described.

  Articles that describe people's experiences are called “experience articles”. The metadata represented by the character string indicating what experience the experience article represents is referred to as “experience attribute” (example of experience attribute: “hamburger store A visit”).

  “Factors” are events, conditions and conditions that have had a positive impact on people's decision to implement experiences.

  In the experience article set with experimental attributes, among the words appearing at least once, the word representing the factor with high probability (with high probability in terms of how the word is used in various sentences) without depending on the appearance context. "Factor word". For example, as a factor word, the word “secondary party” is a factor word in the corpus because it has a high probability of representing a factor when the author visits the store in an experience article about a restaurant visit.

  The attribute of the factor word is “factor attribute”. As an example of the factor attribute, the word “secondary party” can be given a factor attribute of “issue” because it represents an issue when visiting the store among the factors in an experience article about “visiting a restaurant”. Further, the word “stroller” represents an attribute of a visitor when visiting the store, so that a factor attribute of “social environment” can be given.

  The “competitiveness” is calculated based on the similarity of factors between experience attributes.

  In marketing, it is important to understand your company and other companies that compete with your company's products. It is important to understand not only the competitors, but also the viewpoints (for example, any of the five viewpoints of physical environment, social environment, time, issues, and prior state) and specific competitive factors in that viewpoint. It is. For example, subject A (eg restaurant A) and subject B compete for social environment (salaryman group) and time (around 20:00), and subject A and subject C have social environment (housewife group) and issues (girls) To understand competing factors such as competing. At that time, in order to facilitate the understanding of users, each viewpoint does not overlap and is divided by an appropriate number of viewpoints by a classification system structured based on the framework of marketing analysis (Requirement 1). desirable.

  As a technique for realizing the above, a method using a Web document can be considered.

  In the method using a Web document, for example, when there is an experience article set described for each target on the Web, it is possible to determine that the degree of competition is higher between objects whose appearance word frequency distribution has a higher similarity. A cosine similarity or the like can be used as an index for measuring the height of the similarity.

  Furthermore, the appearance words are clustered in advance from the closeness of meaning and co-occurrence of the appearance words, and the degree of competition is determined for each cluster from the similarity of the word appearance distribution assigned to the corresponding cluster for each cluster. Can be calculated. At this time, each cluster is divided according to some criteria, and if the clusters are properly divided (in accordance with Requirement 1), what kind of viewpoint the user is competing with, and its specific competing factors ,I can understand.

  As an example of the clustering algorithm, a method of clustering words having similar meanings as in Patent Document 1 can be considered. In Patent Document 1, a vector representing a semantic expression is assigned to a word in advance, and word clustering is performed based on the vector.

JP 2011-39977 A

  In the conventional technique, in extracting the competition object for each viewpoint in the competition extraction for each viewpoint, when the words are clustered in advance for each viewpoint, they cannot always be separated in a form satisfying the above requirement 1. Therefore, the user cannot easily grasp the viewpoint competing with each competing target.

  Even if a clustering method that considers the meaning of a word typified by Patent Document 1 is used, the meaning vector given in advance does not necessarily have a structure based on the framework of marketing analysis. It is not necessarily divided, and the user cannot easily understand the nature of the set. Therefore, if the similarity score of the appearance word distribution in the set is high, the user cannot easily understand in what viewpoint it is competing. Also, the method of providing the semantic vector does not show a method that satisfies the requirement 1.

  Further, the conventional method does not consider the unexpectedness of the acquired competitive information. The acquired competitive information reflects the opinions of the people who described the Web document. However, it is considered that information that is natural and unexpected information are mixed, and conventional methods cannot distinguish them. For users such as marketers, surprising information is important because it can give new insights, but marketers need to identify the information themselves, which is labor intensive.

  The present invention has been made in view of the above points, and the five factors of the physical environment, social environment, time, issue, and prior state, which represent the events that have had a positive impact on the decision to perform human experience. An object of the present invention is to provide a competitive experience attribute display apparatus, method, and program that display characteristics of competitive brands and competitive elements with factor attributes of viewpoints, and allow users who perform marketing analysis to efficiently understand the competition targets. .

In order to solve the above problems, the present invention (Claim 1) is a competitive experience attribute display device for analyzing a competitive brand from document information,
A factor attribute, which is metadata representing the type of factor, is input to a factor word that represents an event that has had a positive impact on a person's decision to conduct an experience. Based on the factor attribute table in which the assigned data is stored, the experience attribute and factor word, which is metadata represented by a character string indicating what experience the experience article represents, from the experience article set Factor attribute extraction means;
Experience attribute factor word aggregation means for counting the factor word appearance frequency for each experience attribute;
Factor topic model creating means for creating a topic model from the factor word appearance frequency for each experience attribute for each factor attribute;
A degree-of-competition calculation means for calculating the degree of competition between the experience attributes using the similarity of the factor topic distribution for each factor attribute between the experience attributes, and storing it in the competition degree table;
Competitive experience attribute display means for displaying the competitive experience attribute of each experience attribute and related information based on the competitiveness between the experience attributes stored in the competitiveness table.

  In addition, the present invention (claim 2) further includes an unexpected degree calculating means for calculating, as an unexpected degree, a positive / negative inversion value of the similarity degree of the meta experience attribute between the experience attributes obtained by further abstracting the experience attribute.

  Further, according to the present invention (Claim 3), for each experience attribute (hereinafter referred to as “input experience attribute”) input by the user, other experience attributes (hereinafter referred to as “competitive experience attribute”). Among the competitive experience attribute groups with the highest linear sum of the competition level for each factor attribute (hereinafter referred to as “competition level between input experience attribute and competitive experience attribute”), based on the competition level for each factor attribute It further has a competitive experience attribute display means for displaying a radar chart for displaying the calculated value and highlighting characteristic factor words having a high appearance frequency common to the competitive experience attribute and the input experience attribute.

  As an effect of the present invention, when a user such as a marketer grasps a competing object, “there is no overlap, an appropriate number of viewpoints, and each divided by a classification system structured based on the framework of marketing analysis. For each “point of view”, the competing target can be understood, and the specific competing factors can be easily understood.

  Furthermore, by using the unexpectedness index, it is possible to browse well-known competitive relationships and unexpected competitive relationships, and reach the desired information with little effort when requesting either type of information. be able to.

  With the above effects, a system that visualizes competitive experience attributes for specific experience attributes as shown in FIG. 21 can be created, and marketers can intuitively browse the competitive information.

It is a block diagram of the competition experience display apparatus in one embodiment of this invention. It is an example of the factor word table T1-1 in one embodiment of this invention. It is an example of the factor attribute table T1-2 in one embodiment of this invention. It is a program of the factor word acquisition process in one embodiment of this invention. It is an example of the experience article table T2 in one embodiment of this invention. It is an example of the experience attribute table T3-1 in one embodiment of this invention. It is an example of meta experience attribute table T3-2 in one embodiment of this invention. It is an example of the experience attribute factor word table T4-1 in one embodiment of this invention. It is an example of the experience attribute factor word table T4-2 in one embodiment of this invention. It is an example of the factor topic table T5-1 in one embodiment of the present invention. It is an example of factor topic table T5-2 in one embodiment of the present invention. It is an example of factor topic table T5-3 in one embodiment of the present invention. It is an example of the competition degree table T6 in one embodiment of this invention. It is an example of the unexpectedness table T7 in one embodiment of this invention. It is a flowchart of operation | movement of the competition experience attribute display apparatus in one embodiment of this invention. It is a detailed flowchart of experience attribute factor word extraction processing S100 in an embodiment of the present invention. It is a detailed flowchart of factor topic model creation processing S300 in one embodiment of the present invention. It is a detailed flowchart of competition degree calculation process S400 in one embodiment of this invention. It is a detailed flowchart of the unexpected degree calculation process S500 in one embodiment of this invention. It is a figure for demonstrating the process of the unexpected degree calculation part in one embodiment of this invention. It is an example of a display of the competition experience attribute display part in one embodiment of the present invention.

  Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a configuration diagram of a competitive experience attribute display device according to the first embodiment of the present invention.
The competitive experience attribute device in the figure is roughly divided into an input unit 05, an experience attribute factor word extraction unit 10, an experience attribute factor word aggregation unit 20, a factor topic model creation unit 30, a competition degree calculation unit 40, and an unexpected degree calculation unit. 50, a competition experience attribute display unit 60, a recording unit 70, an input external device 80, and an output external device 90.

  The recording unit 70 is a storage medium such as a hard disk and stores a plurality of tables shown below.

  First, each table of the recording unit 70 will be described.

  In FIG. 1, tables T1-1, T1-2, T2, T3-1, and T3-2 (referred to as table group A) are tables in which values are stored in advance before starting the apparatus. In the processing of this apparatus, the value is unchanged. Note that the data structures of the tables T2, T3-1, and T3-2 are defined based on a data structure used in a Web service generally called a review site.

  On the other hand, tables other than the table group A are tables whose values are stored or updated by the processing of this apparatus, and are referred to as a table group B.

  Each table will be described below.

<Factor word table T1-1>
The factor word table T1-1 is a table shown in FIG. 2, and is a table for designating which factor attribute ID each factor word has.

<Factor attribute table T1-2>
The factor attribute table T1-2 is a table shown in FIG. 3, and is a table in which each factor attribute is specified together with its ID.

  Especially in this device, the attribute defined by the classification of the consumer scene classified by Belk (Ref. 1: “Belk, RW: Situational variables and consumer behavior, Journal of Consumer Research (1975))” is used as the factor attribute. That is, the factor attribute table T1-2 has the following five attributes.

1. Physical environment (tangible characteristics that make up the consumer scene)
2. Social environment (others)
3. Time (temporal characteristics of the scene)
4). Challenges (a specific purpose or goal that consumers have)
5. Leading state (the mood or condition that the consumer brought into the scene at that time)
The attribute defined by Belk is used to separate the viewpoints of competition in a way that satisfies requirement 1 above.

  The list of each factor word appearing in the factor word table T1-1 and each factor attribute assigned to each factor word in the factor attribute table T1-2 are as follows in the factor word extraction processing device (not shown). It is assumed that the method has been given in advance.

● Factor word extraction processing The factor word extraction processing will be described below. (Note that among the words that appear more than once in the experience article, the words used to indicate that the author has conducted experience are referred to as “experience words”. ”).

  The factor word extraction process is roughly divided into an experience seed input process S′100, an experience acquisition process S′200, and a factor word acquisition process S′300. Each process will be described in detail below.

S′100) Experience Word Seed Input Process In the experience word seed input process, an input of an experience word from a user is accepted (a few experience words that the user can conceive are assumed). When the user finishes inputting experience words, input of threshold values θ ₁ , θ ₂ , θ ₃ , θ ₄ used in experience word acquisition process S ′ 200 is requested. Then, the experience word list V = {v _i } and θ ₁ , θ ₂ , θ ₃ , θ ₄ input by the user are passed to the experience word acquisition process S ′ 200.

S′200) Experience word acquisition process The experience word acquisition process refers to the recording unit with the experience word list V and θ ₁ , θ ₂ , θ ₃ , θ ₄ passed from the experience acquisition process S ′ 200 as inputs. , A process for acquiring (outputting) an experience word, which includes a process S′201, a process S′202, a process S′203, and a process S′204. Details of each process are shown below.

S'201) Refer to each word table of experience articles and create a list of all words z _{i that} appear more than once (in this case, each word is a unique word when the original label / part of speech pair is unique) Counting, <go, verb-independence> and <go, verb-independence> are counted as separate words).

Among words z _i , a list of words a _i whose part of speech is “noun-sa-variant connection” or “verb-independence” is created.

S′202) The next score p ₁ (a _i ) is calculated for all words a _i .

Part(a_i): 語a_iが各経験記事に出現する際の、直前１語に出現する助詞の頻度分布ベクトル；
part(a_i)の例として、『行く』という単語が経験記事コーパスにおいて、『へ』という助詞を直前１語に取る回数が30回、『に』という助詞を直前１語に取る回数が50回、『と』という助詞を直前１語に取る回数が20回であるとき、part(『行く』)=[30,50,20](ただし左記ベクトルは『へ』、『に』、『と』という助詞を直前に取る回数が格納されるとする)などとすることができる。

Part (a _i ): Frequency distribution vector of particles that appear in the immediately preceding word when the word a _i appears in each experience article;
As an example of part (a _i ), in the experience article corpus, the number of times that the word “Go” is taken as the last word is 30 times, and the number of words “ni” is taken as the last word is 50 Times, when the particle number of “to” is 20 in the last word, part (“go”) = [30,50,20] (however, the left vector is “he”, “ni”, “and” And the number of times the particle is taken immediately before is stored).

cossim (x ₁ , x ₂ ) is a function for calculating the cosine similarity of the input vectors x ₁ and x ₂ as follows.

p₁(a_i)の値が閾値θ₁より高い語b_iのリストを作成する。

A list of words b _i having a value of p ₁ (a _i ) higher than the threshold θ ₁ is created.

S′203) Calculate the next score p ₂ (b _i ) for all words b _i . Note that D (b _i ) is an experience article set including the word b _i .

なお、d_pos(b_i, d)b_iがd内で出現位置が冒頭に近いほど高いスコアが付与される。これは経験記事について、まず経験を実施するまでの過程を記し、その後に経験後のことを記す順番で記述されるモデルを仮定し、さらに、前者の経験を実施するまでの過程(冒頭に近い部分)に経験語がしばしば出現する性質を仮定しているためである。またｄ_pos(b_i,d)の式において、1/2を減算しているのは、単純に出現回数が多い単語のスコアが高くなるのを防ぐためである。

It should be _noted that a higher score is given as d _pos (b _i , d) b _i is closer to the beginning in d. This is an experience article that first describes the process up to the implementation of the experience, then assumes a model that is described in the order in which the post-experience is described, and then the process up to the implementation of the former experience (close to the beginning) This is because it is assumed that the experience words often appear in (part). The reason why 1/2 is subtracted in the expression d _pos (b _i , d) is simply to prevent the score of a word having a large number of appearances from increasing.

After calculating the next score p ₂ (b _i ) for all the words b _i , a list of words c _i whose p ₂ (b _i ) value is higher than the threshold θ ₂ is created.

S′204) The next score p ₃ (c _i ) is calculated for all words c _i .

df(c_i)：c_iの経験記事コーパスにおける出現ドキュメント数；
p₃(c_i)の値が閾値θ₃より高い語のリストを作成し、そのリストを経験語リストとして、要因語獲得処理Ｓ'３００に渡し、処理を進める。

df (c _i ): Number of documents appearing in the experience article corpus of c _i ;
A list of words having a value of p ₃ (c _i ) higher than the threshold value θ ₃ is created, and the list is given as an experience word list to the factor word acquisition process S′300, and the process proceeds.

  As an output example of the experience word acquisition process S′200, when the following experience word seed is given as an input, an experience word seed having the following label can be output.

-Input experience word seed example:
"Go", "Visit"
-Output experience word seed example:
`` Go '', `` Stop by '', `` Stop '', `` Visit '', `` Line up '', `` Line up '', `` Go to '', `` Get to '', `` Visit '', `` Use '', `` Visit '', `` Visit ''
S'300) Factor word acquisition process In the factor word acquisition process, all information is acquired by referring to the experience article word table.

Next, in each experience article d, the following processes (1), (2), and (3) are performed.
(1) For each word appearing in d, for all words whose original form matches any of each e _j , all the appearance positions w ₁ (number of words from the beginning) are acquired.

(2) Next, for each w ₁ , the word located in front of w ₁ is
{And, as, because, so, because}
If either to acquire all of its occurrence position w _2.

(3) Next, for each w ₂ , the word positioned immediately before w ₂ is recorded in the factor word list.

(1) (2) after carrying out all of the experience article processing of (3), and outputs all the words f _i which appeared more theta ₄ times a factor word list as a factor word list.

  The processes (1), (2), and (3) can be described in the form of a program as shown in FIG.

At this time, the set of experience articles d is D, the number of words in each experience article d is size (d), the g-th word in experience article d is w (d, g), the set of experience words is E, and the factor particle is
J = {And, as, so, so, for}
The following processing is executed. Note that count (w _i ) is a process of counting the number of appearances of the word w _i .

In the factor word acquisition process (S'300), focusing on language patterns including experience words and factor particle words, the number of occurrences of the word "" is counted when there is the following expression, and the words counted more than a certain number of times Is recorded as a factor trivial.
(Example 1) Visited at “Secondary Party”.
(Example 2) Used for "Birthday celebration".

  In the factor word acquisition processing, in order to reduce errors and acquire a large number of factor trivial words, the connection with an experience word and a specific factor particle word is used.

● Each factor attribute assigned to each factor word As for each factor attribute of each factor word, it is assumed that one most appropriate factor attribute is determined in advance based on the definition of Belk.

<Experience article table T2>
As shown in FIG. 5A, the experience article table T2 is composed of an experience article ID field, an experience article text field, an experience attribute ID field, and a meta experience attribute ID, and is a table representing the contents of each experience article and its attributes. is there. The experience article text field stores the contents of experience articles as shown in FIG. The experience attribute ID field is a field for designating one experience attribute of the experience article. The meta experience attribute ID field is a field for designating one or more meta experience attributes of the experience article.

<Experience attribute table T3-1>
As shown in FIG. 6, the experience attribute table T3-1 stores each experience attribute together with its ID. The experience attribute refers to metadata represented by a character string indicating what experience the experience article represents.

<Meta experience attribute table T3-2>
In the meta experience attribute table T3-2, as shown in FIG. 7, each meta experience attribute is stored together with its ID. The meta experience attribute is information given in advance by further abstracting the experience attribute. For example, the meta experience attribute of the experience attribute “visit to hamburger store A” is “visit to hamburger store”.

<Experience attribute factor word table T4-1>
In the experience attribute factor word table T4-1, as shown in FIG. 8, when each factor word appears in the review article described for each experience attribute, each <experience attribute ID, factor word> pair is stored. Is done. When the same factor word appears two or more times in a certain experience attribute, lines are added as many times as that number.

<Experience attribute factor word table T4-2>
As shown in FIG. 9, the experience attribute factor word table T4-2 is a table for storing the total number of appearances of each factor word in all articles having each experience attribute ID.

<Factor topic table T5>
The factor topic table includes a factor topic table T5-1 (FIG. 10), a factor topic table T5-2 (FIG. 11), and a factor topic table T5-3 (FIG. 12).

The factor topic table T5-1 shown in FIG. 10 uses a set of appearance frequencies of each factor word w in each experience attribute a for each factor attribute i as an information source, and a latent topic model (Reference 2: “T. Hofmann .: Probabilistic Latent Semantic Indexing. In Proceedings of the 22nd Annual ACM Conference on Research and Development in Information Retrieval, pp. 50-57 (1999)))
p (z | i), p (w | z, i), p (a | z, i)
This is a table for storing p (z | i) when is created. Note that p (z | i) is intended to represent the appearance probability of topic z in the factor attribute i. p (w | z, i) is intended to represent the appearance probability of the factor word w in the factor attribute i and topic z. p (a | z, i) is intended to represent the appearance probability of experience attribute a in topic attribute z and topic z.

  The factor topic table T5-2 shown in FIG. 11 is a table for storing p (a | z, i) when the latent topic model is created for each factor attribute i. Only a table is generated.

  The factor topic table T5-3 shown in FIG. 12 is a table for storing p (w | z, i) when the latent topic model is created for each factor attribute i. Only a table is generated.

<Competition degree table T6>
As shown in FIG. 13, the competition degree table T6 is a table for storing the competition scores between all the experience attributes for each factor attribute i, and as many tables as the number of factor attributes i are generated.

<Unexpectedness table T7>
As shown in FIG. 14, the unexpectedness table T7 is a table for storing unexpected scores among all the experience attributes for each factor attribute i, and as many tables as the number of factor attributes i are generated.

  The operation of the competitive experience attribute display device of the present invention will be described below.

  FIG. 15 is a flowchart of the operation of the competitive experience attribute display device according to the embodiment of the present invention.

  In this system, input processing S000, experience attribute factor word extraction processing S100, experience attribute factor word aggregation processing S200, factor topic model creation processing S300, competition degree calculation processing S400, unexpectedness calculation processing S500, competition experience attribute display processing S600.

S000) Input processing:
First, the input unit 05 waits for the number of topics K for acquiring a competitive relationship and the experience attribute name for displaying the competitive relationship. When the user inputs the former from the input external device 80, the process proceeds to S100. (The number of topics K is used after S300.) When the user inputs the latter, the process proceeds to S600.

S100) Experience attribute factor word extraction processing:
The experience attribute factor word extraction unit 10 receives a set of all articles and factor words as input and outputs a pair <experience attribute ID, each factor word> as a pre-processing of the subsequent processing.

  The detailed processing of S100 is shown in FIG.

  S110) The experience attribute factor word extraction unit 10 first refers to the factor word table T1-1 and the experience article table T2, and acquires all information. Let F be a set of factor words acquired from the factor word table T1-1.

  Next, the experience attribute factor word extraction processing performs the following processing (S121 to S123) for each experience article ID.

  S121) Morphological analysis is performed on the text of each experience article d acquired from the experience article table T2, and a set of appearance words (original form) is acquired.

  S122) For each occurrence word (original form) of each experience article d, the pair <experience attribute ID, each occurrence word (original form)> is sequentially assigned to those that are the same as one of the elements of F. 1 is stored. The “prototype” refers to each appearance word stored in the table.

  S123) When all the experience articles are processed, the process is passed to S200. If there is an unprocessed experience article, the process returns to S121 and proceeds to the next experience article.

S200) Experience attribute factor word aggregation processing:
The experience attribute factor word aggregation unit 20 inputs a pair <experience attribute ID, factor word> set as an input to output data representing the characteristics of the appearance tendency of the factor of each experience attribute <experience attribute ID , Factor word, appearance count> set. Here, the “factor word” is the same as the “appearance word” in S122.

  In S200, first, all information of the experience attribute factor word table T4-1 is acquired. Next, the number of appearances of each factor word (number of lines in the table) is counted for each experience attribute. When the process of counting the number of appearances of each factor word is completed for all experience attributes, all the triples <experience attribute ID, factor word, number of appearances> set are stored in the experience attribute factor word table T4-2. When all the storage is completed, the process is passed to S300.

S300) Factor topic model creation processing:
For each factor attribute, the factor topic model creation unit 30 compresses the dimension of the appearance frequency vector (created in S200) of each factor word in each experience attribute, as a triple <experience attribute ID, factor word, appearance count > Set and pair <Factor word, factor attribute ID> Set as input, appearance probability of each factor topic and factor attribute for each factor attribute, appearance probability and factor attribute for each factor topic for each factor attribute The appearance probability of each factor word for each factor topic is output.

  The flowchart of S300 is shown in FIG.

  S310) The factor topic model creation unit 30 first acquires all the information of the factor word table T1-1, the factor attribute table T1-2, and the experience attribute factor word table T4-2.

S320) Next, for each factor attribute 1 to 5 (assuming i), when a potential factor topic is assumed, the appearance probability of the factor topic z, and each factor word w for each factor topic z Determine the appearance probability of each experience attribute a. That is, p (z | i), p (a | z, i), and p (w | z, i) are determined for all z, a, and w for each experience attribute i. (Remarks: K (1 + M + N) parameters need to be estimated, K: number of factor topics, M: number of experience attributes, N: number of factor words for factor attribute i)
In the factorial topic model creation process, p (z | i), p (a | z, i), p (w | z, i) are used, using the maximum likelihood method to set potential factor topics in a probable manner. Ask for.

  Details of S320 will be described below.

  In S320, the processing of the factor attributes 1 to 5 (i) is repeated. The repeated processing is described below.

  S321) The following values are calculated for every experience attribute. n (a, i, w) is the number of appearances of each factor word w in the experience attribute a. Further, the sum of n (a, i, w) is calculated, and the following value is calculated. n (a, i) is the total number of occurrences of each factor word in experience attribute a.

  S322) p (z | i) = 1 / K is substituted as an initial value for all z. Random values are assigned to P (a | z, i) and p (w | z, i) for all factor topics z, factor words w, and experience attributes a.

  S323) Using the EM algorithm,

を最大化するp(z|i), p(a|z, i), p(w|z, i)を求める。

Find p (z | i), p (a | z, i), p (w | z, i) that maximizes.

  S324) The obtained p (z | i) is in the factor topic table T5-1, p (a | z, i) is in the factor topic table T5-2, and p (w | z, i) is in the factor topic table T5-. 3 is stored.

S400) Competitiveness calculation processing:
The competition degree calculation unit 40 inputs the appearance probability of each factor topic and the appearance probability of each experience attribute for each factor topic for each factor attribute in order to acquire various types of competitive relationships. The degree of competition between each experience attribute is output for each factor attribute.

  A flowchart of the competition degree calculation process is shown in FIG.

  Details of each process will be described below.

  S410) First, all information is acquired from the factor topic tables T5-1 and T5-2.

  S420) Next, the degree of competition between each experience attribute a is calculated for each of the factor attributes 1 to 5 (assuming i).

  Details of S420 are as follows.

  S421) For all factor topics z for all experience attributes a, the value of p (z | a, i) is calculated using the Bayes formula. p (z | a, i) means the belonging probability of each experience attribute to each factor topic. The value of p (z | a, i) is calculated by the following formula.

Ｓ４２２） 全ての経験属性a_jについて、他の各経験属性a_kとの競合度を表す下記のスコアを算出する。なお、ここでは、θ_jiは、経験属性a_jの各要因トピック（要因属性iにおける）への帰属確率p(z|a_j, i)の集合を表すベクトルを意味し、θ_kiは、経験属性a_kの各要因トピック（要因属性iにおける）への帰属確率p(z|a_k, i)の集合を表すベクトルを意味する。下記式は、θ_jiとθ_kiの類似度を測っている。

S422) For all the experience attributes a _j , the following score representing the degree of competition with each other experience attribute a _k is calculated. Note that here, θ _ji means a vector representing a set of attribution probabilities p (z | a _j , i) to each factor topic (in factor attribute i) of experience attribute a _j , and θ _ki is experience membership probability p to each of the factors topic (in factor attribute i) of the attribute _{a k (z | a k,} i) refers to a vector that represents a set of. The following equation measures the similarity between θ _ji and θ _ki .

Ｓ４２３） Ｓ４２２で算出した全スコアを、競合度テーブルＴ６に格納する。
Ｓ５００） 意外度算出処理：
意外度算出部５０では、意外な競合関係を利用者に優先的に見せるために、経験記事テーブルＴ２からペア＜経験属性ID、メタ経験属性ID＞集合を入力として、各経験属性間の意外度を出力する。

S423) All the scores calculated in S422 are stored in the competition degree table T6.
S500) Unexpectedness calculation process:
The unexpectedness calculation unit 50 uses the pair <experience attribute ID, meta experience attribute ID> set from the experience article table T2 as an input in order to show the unexpected competitive relationship to the user preferentially. Is output.

  A flowchart of S500 is shown in FIG.

  S510) The unexpectedness degree calculation unit 50 refers to the experience article table T2 and acquires all information.

  S520) Using the information acquired in S510, the number n (a, s) of co-occurring meta experience attributes s is calculated for each experience attribute a (FIG. 20A). In addition, the total number of co-occurrence meta experience attributes n (a) is calculated.

S530) In S531 and S532, first, by using the co-occurrence information of each meta experience attribute s and each experience attribute a, each experience topic y that abstracts the relationship between each experience attribute and each meta experience attribute is converted into a topic model. Use to create. (FIG. 20B, appearance probability p (y) of y, appearance probability p (a | y) of a in y, appearance probability p (s | y) of s in y are calculated.)
The creation process is performed according to the following procedure.

  S531) p (y) = 1 / K is stored as an initial value for all y, s, and a. Further, random values are stored in P (a | y) and p (s | y).

  S532) Using the EM algorithm,

を最大化するp(y), p(a|y), p(s|y)を求める。

Find p (y), p (a | y), and p (s | y) that maximize.

  S540) Next, for all experience topics y for all a, calculate the value of p (y | a) (the following formula, p (y | a) is the probability of belonging to each y of each a means).

Ｓ５５０） 次に、全ての経験属性a_jについて、他の各経験属性a_kとの意外度を表す下記のスコアを算出する。（なおここでは、ρ_jは、経験属性ρ_jの各経験トピックyへの帰属確率p(y|a_j,)の集合を表すベクトルを意味し、ρ_kは、経験属性a_kの各yへの帰属確率p(y|a_k)の集合を表すベクトルを意味する。下記式は、ρ_jとρ_kの類似度の正負を判定させた値を算出している。メタ経験属性の傾向が異なる経験属性同士ほど、競合関係にあるとき意外であるという仮説に基づく。）

S550) Next, for all the experience attributes a _j , the following score representing the unexpectedness with each other experience attribute a _k is calculated. (In this case, ρ _j means a vector representing a set of membership probabilities p (y | a _j ,) of the experience attribute ρ _j to each experience topic y, and ρ _k represents each y of the experience attribute a _k . This means a vector that represents the set of membership probabilities p (y | a _k ) .The following formula calculates a value that determines whether the similarity between ρ _j and ρ _k is positive or negative. Based on the hypothesis that experience attributes with different values are unexpected when they are in a competitive relationship.)

Ｓ５６０） Ｓ５５０で算出した各経験属性間の意外度を表すスコアを、意外度テーブルＴ７に格納する。
Ｓ６００) 競合経験属性表示処理：
競合経験属性表示部６０では、記録部７０に蓄えられた経験属性間の競合度（競合度テーブルＴ６）や意外度（意外度テーブルＴ７）についての情報を、利用者に分かりやすく示すための処理を行なう。

S560) A score representing the unexpectedness between each experience attribute calculated in S550 is stored in the unexpectedness table T7.
S600) Competitive experience attribute display processing:
In the competitive experience attribute display unit 60, a process for easily displaying information about the degree of competition (competition degree table T6) and the unexpected degree (expected degree table T7) between the experience attributes stored in the recording unit 70 to the user. To do.

  When the experience attribute names are input from the input external device 80 by the user, the competitive experience attribute display unit 60 sorts the factor sums in order of increasing linearity of the degree of competition and the degree of unexpectedness for each factor attribute. The radar chart showing the degree of competition for each factor attribute for each upper experience attribute and the TFIDF upper factor word for each upper experience attribute are displayed on the output external device 90 via the external device. It is assumed that the number of cases to be displayed is set in advance as the top N cases.

  In the competitive experience attribute display unit 60, as shown in FIG. 21, competitive experience attributes (“hamburger store A”, “cafe B” in FIG. 21) having a high degree of competition with respect to the input specific experience attribute (“cafe A” in FIG. 21). ”,“ Hamburger store B ”) can be output.

  Further, by browsing the competition chart (the pentagonal portion in FIG. 21), the user can understand which viewpoint is highly competitive and which viewpoint is low.

Further, it is possible to know what specific elements are competing by browsing the frequent factor words for each factor attribute (column of physical environment, social environment, etc. in FIG. 21).
Furthermore, by using the slide bar in the upper part of the figure, it is possible to select the index (competitiveness and unexpectedness of each factor attribute) to be emphasized (to increase the weight of the linear sum) when displaying competitive experience attributes. it can.

  The internal processing of the competitive experience attribute display unit 60 is shown below.

When an arbitrary experience attribute a _j name is input from the user via the input external device 80, the competition experience attribute display unit 60 displays each competition degree and unexpectedness with each other experience attribute a _k. (The initial value of the coefficient is 1). Next, the radar chart of each _ak is created and displayed in order from _ak having the highest linear sum. Subsequently, in order from a _k having the highest linear sum, among the factor words appearing in the review articles of each a _k , the upper c (T40 in this embodiment) word having the highest TFIDF value is calculated and output. . It should be noted that words that are also included in the upper c words having a high TFIDF value of a _j can be highlighted.

  Further, when the update button is input, the value of the upper slide bar portion in FIG. 21 (competitiveness and unexpectedness coefficient of each factor attribute when linearly summing) is read, and each competitiveness and unexpectedness is read again. Is calculated, and information on each experience attribute is displayed in descending order of the linear sum.

  Note that the operation of the components of the competitive experience attribute display device shown in FIG. 1 in the above embodiment is constructed as a program and installed and executed on a computer used as the competitive experience attribute display device, or via a network. It can be distributed.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made within the scope of the claims.

05 input unit 10 experience attribute factor word extraction unit 20 experience attribute factor word aggregation unit 30 factor topic model creation unit 40 competition degree calculation unit 50 unexpected degree calculation unit 60 strength experience attribute display unit 70 recording unit 80 input external device 90 for output External device

Claims (7)

A competitive experience attribute display device for analyzing competitive brands from document information,
A factor attribute, which is metadata representing the type of factor, is input to a factor word that represents an event that has had a positive impact on a person's decision to conduct an experience. Based on the factor attribute table in which the assigned data is stored, the experience attribute and factor word, which is metadata represented by a character string indicating what experience the experience article represents, from the experience article set Factor attribute extraction means;
Experience attribute factor word aggregation means for counting the factor word appearance frequency for each experience attribute;
Factor topic model creating means for creating a topic model from the factor word appearance frequency for each experience attribute for each factor attribute;
A degree-of-competition calculation means for calculating the degree of competition between the experience attributes using the similarity of the factor topic distribution for each factor attribute between the experience attributes, and storing it in the competition degree table;
Competitive experience attribute display means for displaying the competitive experience attribute of each experience attribute and related information based on the competitiveness between the experience attributes stored in the competitiveness table;
A competitive experience attribute display device characterized by comprising: 2. The competitive experience attribute according to claim 1, further comprising an unexpectedness degree calculating means for calculating, as an unexpected degree, a value of reversal of the similarity degree of the meta experience attribute between each experience attribute further abstracting the experience attribute. Display device. For each experience attribute (hereinafter referred to as “competitive experience attribute”), the degree of competition for each factor attribute (hereinafter referred to as “competitive experience attribute”) against the experience attribute input by the user (hereinafter referred to as “input experience attribute”) Display a radar chart that displays the value calculated based on the degree of competition for each factor attribute for the group of competition experience attributes with the highest linear sum of "competition between input experience attribute and competition experience attribute" 3. The competition according to claim 1, further comprising a competition experience attribute display means for emphasizing and displaying characteristic factor words having a high appearance frequency common to the competition experience attribute and the input experience attribute. Experience attribute display device. A competitive experience attribute display method for analyzing competitive brands from document information,
The human experience factor attribute extraction means takes an experience article set, which is an article describing people's own experiences, as input, and sets the factor type in the factor word that represents the event that has had a positive impact on the decision of the person's experience Based on the factor attribute table storing the data to which the factor attribute that is the metadata to represent is stored, the experience that is the metadata represented by the character string indicating what experience the experience article represents from the experience article set An experience factor attribute extraction step for extracting attributes and factor words;
Experience attribute factor word aggregation means, an experience attribute factor word aggregation step for counting the factor word appearance frequency for each experience attribute,
Factor topic model creating means for creating a topic model from the factor word appearance frequency for each experience attribute for each factor attribute,
A competition degree calculating means calculates a competition degree between experience attributes using the similarity of the factor topic distribution for each factor attribute between each experience attribute, and stores it in a competition degree table;
A competitive experience attribute display step, wherein the competitive experience attribute display means displays the competitive experience attribute of each experience attribute and related information based on the competitiveness between the experience attributes stored in the competitiveness table;
Competitive experience attribute display method characterized by: The unexpectedness calculation means further performs an unexpectedness calculation step of calculating, as an unexpectedness, a value obtained by reversing the similarity of meta experience attributes between the experience attributes further abstracting the experience attributes. 4. The competitive experience attribute display method according to 4. The competing experience attribute display means is a factor attribute among the other experience attributes (hereinafter referred to as “competitive experience attributes”) for the experience attributes input by the user (hereinafter referred to as “input experience attributes”). The value calculated based on the degree of competition for each factor attribute for the competitive experience attribute group with the highest linear sum of the degree of competition for each element (hereinafter referred to as “the degree of competition between the input experience attribute and the competition experience attribute”) And displaying a radar chart to be displayed, and further performing a competitive experience attribute display step of highlighting and displaying characteristic factor words having a high appearance frequency common to the competitive experience attribute and the input experience attribute. Item 6. The competitive experience attribute display method according to Item 4 or 5. Computer
A competitive experience attribute display program for causing each of the means of the competitive experience attribute display device according to any one of claims 1 to 3 to function.

Images