Relationship between Word-of-Mouth Network and Behavioral Tendency — Preliminary Research for Local Downtown Activation

Considering the conversational transmission of a specific topic in human circles as a word-of-mouth (WOM) network, we enumerate two hypotheses. First, the structure of a WOM network is strongly related to the network members’ behavioral tendency concerning the topic of WOM communication. Second, if new information is acquired by some network members, it will be spread in the network. In this research, to verify these hypotheses, we conduct questionnaire surveys of two communities of students in a university, and we investigate members’ conversational relations on the topic of eating-out destinations. To understand the WOM networks, we propose some indexes—n-density, effective edge rate of word-of-mouth, and others—and the spread rate of new information is evaluated based on the results. The results will be used as preliminary data for activation of downtown Kiryu City, including restaurants.


Introduction
These days, the downtowns in most provincial cities in Japan have lost their bustle, and these circumstances have greatly impaired the regional economy.This is true of Kiryu City in Gunma prefecture, whose shopping streets downtown are an emblem of the city.In order to activate the shopping streets, we are trying to find an effective method of disseminating information to citizens to make them recognize the charm of the shopping streets again.According to our investigation of the population composition of Kiryu City, most shop owners and their customers are aging (see note 1).On the other hand, we also found that there are about 7000 high school students and about 3000 university students in Kiryu City.It means that many teenagers who are free in the daytime are in Kiryu City.Therefore, effective information dissemination for both elders and teenagers should be found.Though the internet is popular in the society of today, we believe that face-to-face word of mouth (henceforth WOM), the traditional way of disseminating information, will be more effective in this case, in which elders and teenagers are the key persons in the community.Therefore, in this research, we examine the capability of the WOM network.
The purpose of our research is to evaluate the effect of WOM in relation to the graph structure of a WOM network.First, we evaluate the relationship between the behavioral tendency of the members of the network and the WOM graph structure.Next, we evaluate the spread of new information based on the WOM graph structure.The understanding of WOM in a student community will be used as preliminary data for the activation of the shopping streets of Kiryu City.

Related Research
Problems concerning interpersonal relationship networks are addressed in many fields-such as informatics, sociology, psychology, business administration, and economics-from different viewpoints.We introduce related studies in the following two areas: interpersonal relationship networks and WOM marketing, which is an effect of those networks.study found that some factors are better indicators of social connections than others and that these indicators vary between user populations.

WOM Marketing
In the field of marketing, WOM is examined chiefly in comparison with the mass media influence on an individual's decision making.Rogers (2003) compares advertisement to WOM, and shows that the former has a great influence on an individual at the information-gathering stage but the latter has a big influence at the decision-making stage.Herr et al. (1991) investigated the effects of WOM communications and specific attribute information on product evaluations and found that a face-to-face WOM communication was more persuasive than a printed format but the effect was reduced or eliminated when a prior impression of the target brand was available from memory or when extremely negative attribute information was presented.

The Meaning of WOM Network in This Paper
WOM was defined by Arndt (1967) as oral person-to-person communication between a receiver and a communicator whom the receiver perceives as noncommercial, concerning a brand, product, or service.Brown et al. (1987) added that "the communicator and the receiver have a social tie." In this study, we use the definition of WOM mentioned above.It can be expected that the networks differ depending on the topic of the WOM.Therefore, as the social tie of the communicator and the receiver, we use the existence of conversation concerning meals in daily life as a WOM network.We investigate how information about the shopping streets spreads on this network.In this study, we check just two points.The first point is whether communicators and receivers are interested in the topic.The second point is whether each constituent of the community has a chance to have conversations exchanging information concerning the topic.Meals are a topic of concern for all students, so the network satisfies both these points.Furthermore, the network about meals can be expected to function as a WOM network concerning the shopping streets.

The Hypothesis to Be Examined
In this study, we survey the network of interpersonal relationships that exists in the real community, analyze its structure, and investigate its relation to the behavioral tendency of the constituent members.As a result, we examine the utility of WOM as an information dissemination method to activate the shopping streets.Concretely, a questionnaire survey was designed to examine the following hypothesis: First, there is a strong relationship between the structure of a WOM network and similarity of the members' behavior concerning the topic of the network.Second, when new information is acquired by some members of the network, it will spread in the network.The usability of WOM as an information dissemination method for shopping streets activation is evaluated by assessing the degree of information spreading in a WOM network of a student community.

The Questionnaire Survey
Questionnaire surveys with the same content were administered to 189 junior students at the Department of Computer Science (henceforth C 1 ) and the Department of Mechanical System Engineering (henceforth C 2 ) of Gunma University two times, in April and July.We selected junior students as the investigation object because the groups of students are clear and easy to study.Moreover, steady interpersonal relationship networks are expected for these communities because the students have been in the same department for two years.The analysis was based on the students whose answers were effective in both investigations, and the sample size is 53 for C 1 and 58 for C 2 .The data of C 1 from the first and second investigations are abbreviated as C 11 and C 12 , respectively.The data of C 2 from the first and second investigations are abbreviated as C 21 and C 22, respectively.
The survey carried a signature, and its content is as follows: 1) Eating style at the university in the daytime.
2) The names of other students that the student talked with or emailed about which restaurant to go to for meals, excluding drinking parties, in the preceding week.
3) The name of the restaurant that became the topic.4) Living conditions, such as where the student's home is and whether he or she has a car.

5) Whether does he or she recognize the restaurant Basho (see note 2).
After the first investigation, information about the restaurant Basho was given in a lecture to some of the investigated students of C 1 .Then, the appearance of the spread of the information was observed in the second investigation (Figure 1).

The Definition of the Graph Structure of a WOM Network
In this study, we represent a WOM network, using graph theory (for example, Gross et al. (1998)), as a directed graph whose vertexes are persons, and a directed edge from vertex A to vertex B means that person A talked to person B. We define adjacent level for a pair of vertexes in order to indicate the level of information sharing between a pair in the WOM network.Adjacent level of vertexes A and B is defined as the minimum length of paths from A to B or reverse.If there are no such paths, the adjacent level is defined as ∞.An n-path graph is defined as a graph in which all vertex pairs of adjacent level n or less are connected (Figure 2).An n-path graph of a mutually connected graph becomes a complete graph for a large enough n.If a subset of vertexes and their edges becomes a clique-a subgraph that becomes a complete graph-these persons possibly share all the information in n conversations or less.As an index of the information sharing ease of a WOM network, we define the n-density of a network as the ratio of the number of edges in an n-path graph to the number of edges in a complete graph that has the same vertexes as the original graph.Furthermore, we define transitivity and reciprocality as the auxiliary indexes of a directed network.The former is defined as the ratio of the number of shortcut edges to the number of pairs of vertexes that have a path of length 2. It means the ratio of the relations "a friend of a friend is a friend."The latter is defined as the ratio of the number of reciprocal pairs of vertexes to the number of connected pairs.It means the ratio of relations "a friend thinks of me as a friend."That is to say, the allowable conversation frequency means the quality of the information sharing relation.

Characteristics of the WOM Network Graph
WOM networks concerning meals in each department were identified by the two investigations (Figures 3 to 6), and the characteristics of them are summarized in Table 1 and Table 2.We found that all of the 1-densities of the four networks are about 0.06 when we consider the graphs as directed graphs.There are more large cliques (size of 7) and more isolated students in C 1 thanC 2 .Moreover, C 1 has high transitivity and high reciprocality.The students in C 1 divide into several groups that are highly independent of each other and are intimate in the group.Thus, C 1 has a higher transitivity than C 2 , considered as an undirected graph.The reason for the difference between the two departments can be thought of as a result of the different types of study.The students of C 1 have more one-person jobs, such as programming, and discussion, if it exists, is limited to the one-to-one type.However, the students of C 2 are assigned more experiment practice, and they need to collaborate with different students.Therefore, various interpersonal relationships are composed and a uniform network is established.On the other hand, the number of isolated students decreases in the second investigation, compared with the first one.It seems the students who were isolated at the beginning of the semester became related to a group as time went by.

N-Density of WOM Networks
The n-densities of both WOM networks are shown in Figure 7, where the WOM networks are fixed by the second investigation.The horizontal axis of this figure is path length, which means conversation frequency, and the vertical axis is n-density.It can be found that the n-densities become saturated with 7 path lengths in C 2 and 9 path lengths in C 1 , and the n-densities are 0.9 and 0.7 respectively.That is to say, the information can be disseminated to 90 percent and 70 percent, respectively, of constituent members.The figures are about 30 percent and 20 percent when we assume these networks are undirected networks.

Figure 7. Possibility of information transmission
We consider conversations symmetric among receivers and communicators, and we assume that a reverse edge could not be investigated if there is an edge.Therefore, in the following discussion, directed graphs will be treated as undirected graphs if there is no specification.

Relationship between WOM Network and Behavioral Tendency
Table 3 and Table 4 show the results for the method of taking a meal.Main methods for lunch are using the refectory, purchasing a lunch box, and cooking in both C 1 and C 2 .However, for dinner, most students of C 1 cook by themselves, but there are a variety of methods in C 2 , even though cooking is the main method.

Relationship between WOM Network and Meal Characteristics
Principal component analysis of the ratio of the methods of taking a meal was conducted based on the second investigation.Figure 8 shows an individual factor score scatter plot.Most members of the same clique, indicated by the same mark, have positions in the same quadrant in this figure.That shows a tendency for the persons to have the same eating style.This configuration may indicate that students who have a similar eating style composed the clique on a WOM network, or it may indicate that some intimate members have a high possibility of taking a meal together and so their eating styles became nearly the same.It was clarified that a strong relationship existed between the WOM network concerning meals and similarities in the behavior of taking a meal.

Relationship between Method of Taking a Meal and Adjacent Level
Figure 9 shows the average Euclidean distance of lunch behavior and dinner behavior among individuals who are adjacent in the n-path graph of a WOM network.The horizontal axis means the effective range of information exchanging of the adjacent relationship in the network, and the vertical axis is the average behavior distance defined by Where x ik is a kth factor score of eating style of the ith individual in C 1 or C 2 .The average is taken for all edges (i, j) that compose the n-path graph.
There is a high correlation among the lunch behavior of constituent members whom the adjacent level is near in a WOM network.It is especially remarkable in C 1 .However, the correlation of dinner behavior was not clear.It is understood that the influence of conversation at the university on dinner behavior is restrictive.

Change in Recognition by Spreading of Specific Information
Seven students of C 1 and no students of C 2 participated in an information design class between the two investigations, and information concerning a restaurant named Basho was given to them in this class, in order to find the change in recognition caused by information spreading.Basho had once been a topic in the local newspaper, so the WOM network of the students is not the only information source.If some students got to know of Basho, the cause of the change is uncertain.However, it can be determined that the difference in the recognition change between the two departments is the effect of our intervention.Table 5 summarizes the change in recognition of Basho.When we compare the improvement of recognition between the two investigations, the improvement in C 2 , in which there was no intervention, is only about 7%.On the other hand, in C 1 , in which there was positive intervention, the improvement is about 20%.The 13% difference between the two departments is thought to be an effect of the intervention.We now define the effective edge rate of WOM-that is, the rate at which new information spread to the person of adjacency level 1.The effective edge rate of WOM is defined as Ee/E in our case.E is the number of edges from the students who already knew about Basho in the first investigation or participated in the information design class to the students who did not know about it, and Ee is the number of edges that may disseminate the information effectively.That is to say, Ee is the number of edges from the students who already knew about Basho in the first time or participated in the information design class to the students who got to know about it by the time of the second one.In the case of Basho, the index is 23.1% for C 1 , and 7.8% for C 2 .
It is thought that the effective edge rate of a WOM depends on how much the communicator and the receiver are interested in the information to be disseminated.The difference of probability of15.3% is considered an effect of the intervention.We think that the intervention from the class had more impact than the local newspaper and other information sources.

Discussion
In our study, we verified by investigating the WOM networks of the students that there is a strong relationship between the behavioral tendency concerning a topic disseminated on a WOM network and the WOM structure.Moreover, it was confirmed that new information is disseminated in the network if some constituent members of the network acquire it.Therefore, a WOM network was proved as an effective way to disseminate information.
The topic disseminated in the WOM network concerned a restaurant in the center city that has no high special fascination for the students.However, new recognition of it was found as a result of the intervention in the WOM network.Therefore, a new way to increase recognition of the shopping streets could be to make known to some students the fascination of the shopping streets by letting them join some event or cooperate with a shop owner.
Then information about the shop should be expected to be disseminated to the students who are adjacent to the ones who already know about it through a WOM network.That is to say, using the WOM network effectively can be expected to affect the consumption behavior of a young person.In addition, the shopping streets will be bustling with visits of young persons, and that will be a way to create downtown activity.
Only dissemination to students of adjacent level 1 from the students who experienced intervention was observed in our study.We might observe wider dissemination if the topic has a more interesting style.
The following are problems for analysis.First, in this study, all of the observed conversational relations were treated as the same closeness for dissemination of information.However, it is thought that the possibility of information dissemination is greatly influenced by the strength of the interpersonal relationship.Therefore, the next step is to make a model of the effective edge rate of WOM by measuring the conversational relationship quantitatively and the impact of topics.Second, it can be thought that a topic is saturated in the entire WOM network if the effective edge rate of WOM for a topic is high and the WOM network has high density.Thus, another topic is finding the conditions to produce a large prevalence in a WOM network, applying percolation theory.

Figure 1 .
Figure 1.Method of research

Figure
Figure 2. N-Path graph

Figure 8 .
Figure 8. Factor scores of the two department's students methods of taking a meal Note: ▲is G1, ■ is G2, • is G3, ◆ is G4, ▼ is G5, × is G6,and • is the others of Figure 5.
and • is the others of Figure5.The 1st, 2nd, and 3rd principal components are lunch box, cooking, and restaurant, respectively.

Figure 9 .
Figure 9. Relationship between adjacent level and distance of lunch/dinner behavior

Figure 10 .
Figure 10.Information spreading process inC 2 Note: ■represents the students who already knew Basho at the 1st investigation, ▲represents the students who took the class, and •represents the students came to know of Basho by the 2nd investigation.

Figure 11 .
Figure 11.Information spreading process in C 1 Note: ■represents the students who already knew Basho at the 1st investigation, ▲represents the students who took the class, and •represents the students came to know of Basho by the 2nd investigation.

Table 1 .
Basic character of networks

Table 5 .
Change of recognition rate between two investigations -that is, half the number of students who got to know of Basho between the two investigations of C 1 -are not the students who participated in the information design class.It is expected that that is the result of information spreading.The detail of spreading processes is shown in Figures10 and 11.All five students who recognized Basho newly in C 1 are connected with adjacency level 1to students who had already known about Basho in the first investigation or took the information design class.On the other hand, only four students newly recognized Basho in C 2 .Two of them are connected to a student who had already known about Basho in the first investigation.It is highly likely that all of the students who got to know of Basho between the two investigations acquired information about it through the WOM network.