A Computational Model and Empirical Study of Older Adults’ Interaction Behavior with a Mediatized Society

Social media is not a new thing only in modern society, although all kinds of social media have appeared in history, they all have a common basis - sharing between people [24]. Sharing is a natural part of being human, and socializing is also a need. In social networks, humans exchange information with and about others to build relationships, and create a sense of belonging and security by creating a shared environment. Therefore, the use of social media is to establish and strengthen social bonds. From language to text, printing, and today’s development of mobile Internet, the iteration of media technology has created milestones for the innovation and development of social media, and it can be said that needs and carriers are the two key factors that have pushed the emergence of social media from the emergence of social media to the boom.

Nowadays, online social media has created new performance stages for the elderly, who are both actors and spectators, as well as observers and observed. The extensive use of living landscapes further contributes to the landscaping of the real world, where “the landscape as a whole is a ‘mirror image’ of the viewer.” The landscape itself is the performance, and it is as readable as any other text. In modern society, the world is seen more radically as landscape, and thus a kind of gaze and viewing of landscape has become more prevalent in everyday life, a gaze specific to modern society that can be described as a gaze from possession to presentation, a new way of grasping the world as mentioned above, which, together with the power of online social media, has turned the world not only into a collection of people’s personal life landscapes, but also into a a series of privatized performances.

All performances contain a certain degree of rituals and rules, the nature of which is related to the physical and social distance between the performer and the audience.

Focusing on the way electronic media reconfigure the social scene between people, he argues that it is similar to the erection and dismantling of a wall, and that the existence, disappearance and alteration of this wall will have a certain impact on people’s social interactions, in other words, the social behavior between people is shaped and modified according to the social scene in which they find themselves, and that the Internet builds a virtual society for us, which redefines our social scene. So what is the relationship between electronic media and social scenes? The relationship between the two can be discussed through the following ideas: electronic media, change of information flow pattern, and redefinition of the scene.

“The ability of particular social actors to accept each other in a given scene often depends on limited awareness of each other in other scenes.” In other words, it is difficult to react to one scene as if it were another. The separation of social scenes implies the separation of social behaviors, such as going to and from work, going to and from class. This explains why people who know each other well in one scene feel awkward when they meet in another, and the solution to this feeling of awkwardness is to adopt a social behavior that is compatible with both the originally separated social scenes and the now-integrated social scenes. This is what Merowitz calls mid-zone behavior. In order to maintain harmony in the relationship between role and behavior, the old man tries to present some sort of relatively consistent personality of the self to each audience member, so any information the audience member receives about his or her own behavior from other social scenes must be taken into account when the old man performs any of the acts. The use of cell phones has provided a new fusion of virtual reality scenarios for the elderly’s social interactions, and has successfully matched them with new behaviors. Having analyzed “where I am” (i.e., the reassembly of the social performance stage) and “who I am” (i.e., the reconceptualization of social roles), this paper needs to move on to “what I am doing” (i.e., new social behaviors). “(i.e., the emergence of new social behaviors).

Vector space models have a long history, and given this representation, the resulting documents can be ranked for relevance by calculating the distance between the document representation points and the query representation points [25]. In layman’s terms, a similarity measure is used to compute the similarity between a document and a user’s query statement. Among a large number of similarity metrics, the best one is the cosine similarity computation, which calculates the cosine of the angle between the query vector and the document vector [26]: (1)Cosine(Di,Q)=∑j=1tdjj⋅qj∑j=1tdjj2⋅∑j=1tqj2

The following discussion is about the important issue of word weights in vector space modeling. Most of the current word weight calculations are based on the tf*idf framework, where tf represents the word frequency of the word in the document and idf is the inverse document frequency.

tf is used to reflect the importance of the word in the document (query statement), usually counting the number of occurrences of a word in the document is used for calculation. Usually, if a word occurs more frequently, the word is more responsive to the content of the document. The formula for calculating the word frequency factor tf is as follows: (2)tfik=fik∑j=1ifij

where tf_ik is the word frequency of word k in document i and f_ik is the number of occurrences of word k in document i. In the vector space model, the document collection may contain documents of different lengths. A long document may contain a large number of words with fewer occurrences and a small number of words with more occurrences. A large number of retrieval model experiments have confirmed that the problem of word occurrence frequency can be well optimized by taking logarithmic

idf reflects the importance of a word in the document set and. If there are more documents containing a word, then the word is less discriminative for document search, which means that the search using the word is less effective. The inverse document factor idf is calculated as follows: (3)idfk=logNnk

where idf_k is the inverse document frequency of word k, N is the number of all documents, and n_k is the number of documents in which word k appears.

The tf*idf-framework uses the product of the word frequency factor and the inverse document factor as the feature weights, the larger the feature weights, the better this word is likely to be as a search indicator: (4)weight=tf*idf

Probability theory is a powerful basis for demonstrating and calculating uncertainty in the information retrieval process. After decades of exploration, probabilistic retrieval models have become the dominant retrieval models.

In most of the retrieval models document relevance is binary, i.e., for a search there are only two types of documents, i.e., relevant and irrelevant documents. When a new document is added to the document collection, the system will decide which category of documents it exists in based on the description of the document.

The probabilistic retrieval model provides a formula for calculating the probability of document relevance by categorizing documents and adding them to the set of highly probable documents. In other words, if P(R|D) > P(NRID) indicates that a document is relevant and vice versa it is irrelevant, where P(R|D) is the conditional probability of the document’s relevance based on the document description and P(NRID) is the conditional probability of irrelevance. This is known as Bayesian decision rule and the system utilizes Bayesian classification for document categorization.

The probability of an individual belonging to a group of relevant locations is defined as P(RII), the probability of belonging to a group of non-relevant points is defined as P(NRII), and if P(RII) > P(NRII) then the user is relevant to the location of a specific behavior. In order to compute P(RII) and P(NRII), two conditional probability formulas are first expanded according to Bayes’ rule [27]: (6)P(R|I)=P(D|I)P(I)P(I) (7)P(NR|I)=P(I|NR)P(NR)P(I)

where P(R) is the a priori probability of correlation, while P(I) is a normalized constant value. Bringing this into Equation P(RII) > P(NRII), the problem can thus be transformed into calculating the behavioral location correlation through the following equation: (8)P(I|R)P(I|NR)>P(NR)P(R)

where the left side of the equation is referred to as the likelihood ratio. If the likelihood ratio is used as the individual user score, then the top ranked individual users will be the ones with high likelihood in the user population.

Perhaps behavioral locations do not often occur in independent patterns. Assuming p_j is the probability that behavioral location j occurs in the correlated user group, then the probability that an individual user occurs in the correlated location group is p1×(1−p2)×(1−p3)×p4×p5 . (1−p2) is the probability that behavioral location numbered 2 does not occur in an individual user. For factor P(NRII), assuming that s_j is used to denote the probability that behavioral location j occurs in the group of non-relevant locations, then the probability that an individual user belongs to the group of non-relevant locations is s1×(1−s2)×(1−s3)×54×s5 .

With this, the likelihood ratio can be expressed as: (9)P(I|R)P(I|NR)=∏j:Ij=1pjsj⋅∏j:Ij=01−pj1−sj

where Πj:jf=1 denotes the cumulative product of probabilities that individual user I has been to the behavioral location, and Πj:jf=0 in the right part denotes the cumulative product of probabilities that individual user I has not been to the behavioral location. Further, transforming the formula yields: (10)P(I|R)P(I|NR) = ∏j:Ij=1pjsj(∏jIj=11−sj1−pj∏j{Ij=11−pj1−sj)∏j:I,=01−pj1−sj = ∏jIj=1pj(1−sj)sj(1−pj)∏j1−pj1−sj

Since p_j and s_j refer to global probabilities, the second cumulative product Πj does not work. Multiplying a large number of decimals affects the accuracy of the results, so taking the logarithm of the product yields the following scoring formula: (11)∑j:Ij=1logpj(1−sj)sj(1−pj)

In order to avoid inaccuracies in the calculation of the behavioral factors without the above scenario, consider introducing the average number of total time slices of regional users to balance the total time slices of individual users. Let tts_i be the total number of average daily activity time slices for user i, ttst=∑j=1mtstj .

Use avts to denote the total number of average daily activity time slices for all users, avtts=(∑i=1i=n∑j=1j=mtsij)/n . Define the user i activity duration adjustment factor E_i: (12)Ei=k1((1−b)+b⋅ttsiavtts)

E_i represents the consideration of the total number of user i time slices. k₁, b are empirical parameters, and b pairs are used to regulate the user i time-slice ratio, which takes a range of [0, 1] values. When b taking a value of 0 means that the time-slice ratio is canceled, while when b taking a value of 1 means that the full time-slice ratio ttsiavtts is used.

Define the behavior factor: (13)BeFij=(k1+1)⋅tsijEi+tsij

BeF_ij indicates the degree of relevance of user i to behavioral location j; the higher the value, the more relevant the user is to the behavioral location. Finally, BeF_ij*BF_j shows the framework for calculating the weights of users i to behavioral locations j. (14)weightij=BeFij⋅BFj

where the parameter in BF_j, pv_j is the total number of visitors (foot traffic) to the behavioral location j on average per day, i.e., the number of columns j in G that are not zero. BF_j is the global factor, which considers the relative importance of behavioral location j. From Eq. (14), the more time slices ts_ij a user i has for a behavioral location j and the less foot traffic pv_j at that behavioral location, the higher the weight of that behavioral location.

Expand the model equation into full form: (15)weightij=(k1+1)⋅tsijk1((1−b)+b⋅ttsiavtts)+tsij⋅logN−pvj+0.5pvj+0.5

Therefore, taking ts_ij as private data (tts_i can be obtained by ts_ij cumulative calculation), then the remaining necessary factors BF and avtts in the metric model are called public factors. The public factors will be calculated based on the union matrix G, so the union matrix G will be stored in the cloud platform server and the public factors will be calculated based on it.

In order to more comprehensively analyze the behavior of the elderly cell phone using APP, this paper analyzes the mediated social interaction behavior of the elderly in a community from different aspects. The data collected are mainly the two factors of time and location, which include date, APP name, start time of use, end time of use, and GPS location information, respectively. Using the BM25 retrieval model to calculate the above data, the duration and frequency of cell phone use can be again obtained.

The results of the length of time and frequency of use of all cell phone apps over the full period of time collected are shown in Figure 2. It can be seen that the length of time and frequency of using cell phone applications of the elderly both with the increase of cell phone applications into a power law decreasing relationship.

Figure 2.

Power law distribution of APP usage frequency and usage duration

Figure (a) shows the average frequency of using apps per person per month, and Figure (b) shows the average duration (minutes) of using apps per person per month. Figure (a) horizontal coordinate is the number of cell phone apps and vertical coordinate is the average value of frequency of using cell phone apps in a month for a person, and Figure (b) horizontal coordinate is the number of cell phone apps and vertical coordinate is the average duration of using cell phone apps in minutes for a person in a month. When the number of apps is more than 120, the average monthly frequency of the elderly is close to 0. At the same time, when the number of apps used by the elderly is 1, the average monthly average length of use is more than 1600min, and when the number of use is more than 120, the length of use is also close to 0. From this, it can be seen very clearly that, no matter whether it is the length of time of cell phone apps or the frequency of use, it is a power law with the number of cell phone apps decreasing relationship. That is, with the increase in the number of cell phone applications, the duration and frequency of use of cell phone applications decreases exponentially, and the monthly frequency and duration of use of more than 300 APPs shows a power law distribution, indicating that people spend most of their time on very few APPs.

In the statistical results of all APPs used during the full period of data collection by all the elderly people participating in the experiment, the top ten APPs used according to the length of use and frequency of use are shown in Figure 3.

Figure 3.

Use (length and frequency) in the top 10 app

The top ten APPs with the longest duration of use and the top ten APPs with the highest frequency of use are not necessarily one-to-one, but they are generally positively correlated. That is to say, the APPs with the longest usage time are generally ranked ahead in terms of usage frequency, but not absolutely. There are also individual APPs with the longest usage time ranked ahead but lower usage frequency, such as the APPs Zhihu and iReader, with the usage time of 1103.594 and 796.452 respectively, and the usage frequency of only 203 and 98 times.

Figure 4 shows the distribution of the percentage of time spent using social network APPs in different locations during the day, which shows the behavioral pattern of cell phone use among the elderly, the largest percentage of time spent using cell phones within an hour is in community centers, and is in the most used time period of cell phone use from 19:00 p.m. to 12:30 a.m. onwards, with the percentage of time spent in this time period ranging from 0.05 to 0.23. Mostly at the library from 7am to 8am, from 11am to 13pm, and from 17pm to 19pm, yet the length of cell phone use was much less intense than at the community center. A “small peak” of cell phone use occurs in parks and plazas from 1:00 p.m. to 3:00 p.m., with an average percentage of time spent close to 0.1.

Figure 4.

Social networking APPs use time in different locations

Figure 5 shows the behavioral pattern of mobile app usage based on geographic location, the horizontal coordinate of the graph indicates the time and the vertical coordinate indicates the duration of usage, the longer the time of using the mobile app, the higher the vertical coordinate. Each circle indicates the user’s use of the APP, and the area of the circle also indicates the length of time the user has been using the APP, the longer the user has been using the APP, the larger the area of the circle will be. The darker the color indicates that the type of app used is more entertainment oriented and the lighter the color indicates that the type of app used is more social oriented. Additionally, text labels were given to each circle to indicate where the user was during that time period. Figure 5 shows the geographic location and usage of mobile apps by older adults 24 hours a day, including the length of time the app was used.

Figure 5.

Mobile APP usage mode based on geographic location

In Figure 5, the location of the users is at home after 12pm until 7am, and the start of the activity time is from 8am to 12pm and from 2pm to 6pm. It can be out that most of the older people are in the park during the activity time, while using the cell phone less than the non-activity time, and the length of use is not more than 20 min. Other than that, the location with the longest time spent using the cell phone is at home. The locations with the least amount of time spent using the mobile app included: activity centers, neighborhoods, libraries, and parks. Usage at home and in theaters is essentially at the same level, although during the evening period, usage is slightly higher at home than during the daytime. Time period also has an impact on cell phone use, and it is clear that the length of time spent using a cell phone during activity time periods is much shorter than using a cell phone during rest periods. However, the effect of location on cell phone use is very significant, for example, the length of time spent using cell phone apps at home, in the community, and in the library, also during the evening, varies greatly, with the longest length of time spent at home, over 30 min, and no more than 20 min spent in the community and in the library.

Figure 6 shows the distribution of the sample’s average daily length of time using cell phone media, there are differences in the average daily length of time using cell phone media among older people at different stages, the number of older people whose average daily length of time using cell phone media is between 2.1 and 4 hours is the highest among older people in the three age groups of 60-70 years old, 71-75 years old, and 76-80 years old, which are 65, 90, and 30 respectively, with a proportion of respectively 45.14%, 47.12% and 47.62% respectively, and the number of elderly people whose average daily duration of using cell phone media is between 0 and 2 hours is the second highest, with 49, 60 and 20 people respectively, accounting for 34.03%, 31.41% and 31.75% respectively. The number of elderly people who used cell phone media for an average of 4.1~6 hours per day was smaller, with 25, 34, and 11 people respectively, accounting for 17.36%, 17.8%, and 17.46% respectively. The number of elderly people who use cellular media for more than 6 hours a day on average is the least, with 5, 7 and 2 people, accounting for 3.47%, 3.66% and 3.17% respectively. For the number of elderly people over 80 years old decreases with the increase of the average daily time period of using cell phone media. There are 10 elderly people over 80 years old who use cell phone media between 0 and 2 hours per day on average, and 2 elderly people who use cell phone media for more than 6 hours per day on average.

Figure 6.

The average daily use of mobile media hours

Figure 7 shows the purpose of the sample’s exposure to cell phone media, for the purpose of usual exposure to cell phone media, 61.485% of the elderly chose to socialize more conveniently, with friends and relatives using it and leisure and entertainment and killing time coming next, accounting for 55.965% and 47.765% respectively. And the elderly who chose to learn knowledge were the least, accounting for only 9.152%. It can be seen that social convenience is the main purpose for the elderly to contact the cell phone media, and entertainment and leisure applications are the most used applications by the elderly.

Figure 7.

Sample mobile media contact purpose

The dependence level of cell phones was categorized into four levels: no dependence, mild dependence, moderate dependence, and heavy dependence. It was found that different usage patterns of smartphones can affect the degree of smartphone dependence. This paper analyzes and compares the years of use, hours of use, functions and evaluation of use of smartphones with different levels of cell phone dependence, and focuses only on the three levels of dependence: non-dependence, mild dependence and moderate dependence because there is only one person with heavy dependence among the interviewees. The elderly have different levels of cell phone use and dependence, as illustrated in Table 1.

In terms of the number of years of smartphone use, the longer the number of years of use, the higher the relative degree of cell phone dependence. Among those with the degree of non-dependence, nearly half of them have used their cell phones for less than 4 years, among those with mild dependence, there are relatively more people who have used their smartphones for 2-4 years and 4-6 years, among those with moderate dependence, there are more people who have used their smartphones for 4-6 years and 6-8 years, and there is another 10.57% of the respondents have been using their smartphones for more than 10 years, and in terms of the length of daily use of smartphones, the length of cell phone use of those who are mildly dependent and moderately dependent is more than that of those who are not dependent.