This research adopts a quantitative research approach using data obtained from National Bureau of Statistics (NBS) on active voice and internet per state, porting and tariff information. The dataset has a historical information as it comprises of second, third and fourth quarter of both voice and internet records of users for 2021 [14]. The first quarter was for June 2021, second quarter was for September 2021 while the third quarter was for December 2021. The historical attribute makes it possible to study the behaviour of each network provider in a given location service area over a period of time. For each state and network provider (Mtn, Airtel, Globacom and 9Mobile/Etisalat), the voice and data records are tabulated in line with the associated quarter depicting the period in 2021. An aggregation of the dataset for three quarters, Q1, Q2 and Q3 was found to be a total of 37 data points which was indeed small. However, to compensate for this, the affinity propagation machine learning algorithm was employed to cluster the dataset into states that have similar LSPs in terms of voice and internet subscription tariffs. A framework for the location service recommender was thus developed based on the dataset.
2.1 Conceptualizing the LSP recommender
The existing service plan in Nigeria's mobile network are Voice and Internet service plans. There are four major mobile network providers which are Mtn, Airtel, Globacom and 9Mobile/Etisalat. These network providers distribute their service plans to every part of the country in a competitive mode seeking to satisfy their customers through enhanced service delivery. Each of them presents juicy and promising service plans and associated tariff some of which are listed as follows:
a. Mtn
Mtn service plan and associated tariff as obtained from [15] and [16] are summarized in Table 1
Table 1
Mtn Service plans and Tariff
S/No | Tariff | Call Rates | Migration Codes | Emphasis |
1 | Mtn Pulse | 13k/sec + initial 30k/sec daily charge | *123*2*2# | Voice call |
2 | BetaTalk | 75k/sec (N45/M) | *123*2*1# | Voice call |
3 | mPulse | 15.36k/secs | *123*2*3# | Voice call & Data |
4 | XtraSpecial Postpaid | 3.33k/sec | *409# | Voice call & Data/Internet |
5 | Mtn Truetalk | 14kobo/sec + N10 daily access fee | *123*2*6# | Voice call |
6 | Mtn Extra talk | N27/min | *312*2# | Voice call |
7 | Mtn Extra data | N0.23/MB | *312*2# | Data/Internet |
In addition to the listed tariff plans, Mtn provides routine as well as social media data subscription for hourly, daily, weekly and monthly use depending on the amount recharged. For instance, Mtn charges N200 for 2GB worth of data on hourly basis while you can get as much as 1.2GB for N1000 on monthly basis.
b. Airtel
Airtel service plan and associated tariff as obtained from [17] are summarized in Table 2
Table 2
Airtel Service plans and Tariff
S/No | Tariff | Call Rates | Migration Codes | Emphasis |
1 | Airtel Ovajara | N24/Min or 40k/Sec | *544# | Voice call |
2 | Airtel SmartTrybe | N0.78/Min or 13k/Sec | *412# | Voice call & Data/Internet |
3 | Airtel smartTalk | 15k/Sec + N10 on 1st call | *141*505# | Voice call |
4 | Airtel Smart Premier Bundle | 11k/sec for Intl. calls | *470# | Voice call & Data/Internet |
Airtel also has additional routine as well as social media data subscription on daily,, weekly and monthly use depending on the amount recharged. For instance, Airtel charges N50 for 200MB worth of data on daily basis while you can get as much as 1.5GB for N1000 on monthly basis with data bonus for social media attached.
c. Glo
Glo service plan and associated tariff as obtained in [18] are summarized in Table 3
Table 3
Glo Service plans and Tariff
S/No | Tariff | Call Rates | Migration Codes | Emphasis |
1 | Glo Always on | 12k/sec | *301# | Voice call |
2 | Glo Berekete 10X | 36k/sec for calls on main account and 77k/sec for calls on bonus account | *777# | Data/Internet & Voice call |
3 | Glo 11k Per Sec | 11k/sec | *301# | Voice call |
4 | Glo 22X plan | 75k/Sec | *777# | Voice call & Data/Internet |
In addition to the listed tariff plans, Glo also offers attarctive data bundles for routine as well as social media data subscription for daily, weekly and monthly use depending on the amount recharged. The cost for GLO daily data plan is N50 for 50mb, while they charge N1000 for 3.9GB on monthly basis.
d. 9Mobile/Etisalat
MTN service plan and associated tariff as obtained from [19] are summarized in Table 4
Table 4
9Mobile/Etisalat Service plans and Tariff
S/No | Tariff | Call Rates | Migration Codes | Emphasis |
1 | 9konfam | 26k/sec and 78k/Sec after 9 times recharge | *1400# | Voice call & Data/Internet |
2 | Moreflexplus | 59k/Sec | *320# | Voice call & Data/Internet |
3 | Moreclip | 11.26k/Sec and 25.6k/Sec after the first 50 Seconds | *244*1# | Voice call |
4 | Morelife Complete | 11k/Sec on daily access of N5.12k | *620*1# | Voice call |
5 | More Business | N1000/70Mins | *310# | Voice call & Data/Internet |
9Mobile also presents other data plans such as daily charge of N50 for 50Mb of data, N1,200.00 for 6.5Gb worth of data and lots more.
One thing that is clear from the service plans and tariff is that there is bound to be different choice of subscription across the country for each network provider. Nigerians enjoy 2G, 3G, 4G and 5G network coverage technology. However, according to [20], the preferred network coverage technology in Nigeria is still 3G, with 79 per cent penetration. It is also stated that over 40 percent of the population in Nigeria subscribed to 3G mobile broadband in 2022 [21]. Therefore based on [22] and using 3G network primarily, Nigeria's network coverage for the four main network providers is presented in Fig. 1.
Figure 1 shows the dispersion of network coverage for the four major network providers across Nigeria. From the figure, Mtn has more coverage than other providers while 9Mobile have the least coverage. The figure also shows that the four network providers covers the whole country with each having their own subscribers based on the LSP. Notable urbanized states such as Lagos, Rivers, Enugu, Kano, Abuja and others are more densely subscribed across the network providers showing also that urbanization contributes to LSP.
2.2 The LSP recommender model
The LSP recommender model is a 3-tasked module model which are;
Data extraction module, Recommender module and Classifier module
2.2.1 Data extraction module
The data extraction module is the input section of the LSP recommender system. Its main function is to extract the required data for processing. The LSP interfaces with the source database for each network provider. The output of the data extraction module is a pair given as [Vi, Ii] where Vi represents an aggregated output for voice calls and Ii is the aggregated output for internet subscription respectively.
The derivation of Vi and Ii is based on the Nigerian Mobile Telecommunication dataset which are released in quarters annually. The 2021 mobile telecommunication dataset is made up of the number of subscriptions for active voice and internet per state for three consecutive quarters (1st Qtr 2nd Qtr, 3rd Qtr) in 2021. Each of the four network providers were represented in all the 37 states. The layout and aggregation performed on the dataset is shown in Table 5.
Table 5
Aggregation and Layout of the Dataset
Voice Calls & Internet/Data |
| Mtn | Airtel | Glo | 9Mobile |
State | !st Qtr | 2nd Qtr | 3rd Qtr | Ave | !st Qtr | 2nd Qtr | 3rd Qtr | Ave | !st Qtr | 2nd Qtr | 3rd Qtr | Ave | !st Qtr | 2nd Qtr | 3rd Qtr | Ave |
1 | | | | | | | | | | | | | | | | |
. | | | | | | | | | | | | | | | | |
. | | | | | | | | | | | | | | | | |
37 | | | | | | | | | | | | | | | | |
To extract the required Voice data for the LSP recommender system, the average number of voice call subscribers denoted as V_Aveip for the three quarters are computed for each network provider using the equation
\(V\_{Ave}_{ip}= \sum _{i=1}^{n}{Qtr}_{i}\) 1
where i denotes the number of quarters (Qtr) available per annum and it is set of natural numbers; i.e. {1,2,3,…} and p denotes the respective network providers which is also a finite set of natural numbers given as {1,2,3,4 }
Similarly, the average number of Internet subscribers denoted as I_Avei were also computed for each network subscriber across the states using the equation:
\(I\_{Ave}_{ip}= \sum _{i=1}^{n}{Qtr}_{i}\) 2
The aggregation gives rise to two variables for each network provider denoted as;
\(V\_{Ave}_{ip}\) = Voice subscription for any given state (i) and network provider (p)
\(I\_{Ave}_{ip}=\) = Internet subscription for any given state (i) and network provider (p)
2.2.2 Recommender module
The recommender module is the engine house of the LSP model. It is the thinking part which is based on the affinity propagation machine learning model. The recommender module accepts input in pairs from the data extraction module [\({V}_{{Ave}_{ip}}, I\_{Ave}_{ip}\)] as required by the affinity propagation model. Affinity propagation is a non supervised machine learning model used for clustering and identifying similarities among groups. It will be suitable for clustering the states and identifying similarities within them. The authors in [23] employed affinity propagation to develop a hybrid based recommender system. The authors in [24] successfully showed that affinity propagation model was a preferred model for clustering while the authors in [25] employed the affinity propagation model to identify similarities in image processing. The affinity propagation model makes use of similarity matrix computed across all the data points. The affinity propagation model is known to perform very well with small datasets and ensures a lower clustering error.
The affinity propagation model for the LSP recommender system makes use of three basic matrices which are defined as follows:
a. Similarity Matrix(S): This matrix determines the similarities between the mobile telecommunication data points across the 37 states. The similarity matrix computes a single value 'similarity score' based on their features using the negative squared distance between the data points using the equation:
\(S\left(i,k\right)= -{‖x\left(i\right)- x\left(k\right)‖}^{2}\) 3
S(i, k) represents the similarity score between the data point pair, i and k.
\({‖x\left(i\right)- x\left(k\right)‖}^{2}\) is the Euclidean distance between the two data points.
b. Responsibility Matrix(R): The responsibility matrix is used to represent the suitability of a given data point as a cluster center (exemplar) for another data point. It is denoted as
R(i, k), which shows how well the data point 'i'' is well suited as a cluster center for data point 'k'. The responsibility matrix is iteratively computed and updated after each iteration.
c. Availability Matrix (A): This availability matrix A(i, k) represents the 'availability' of a given data point to serve as an exemplar for other data points. The availability matrix ensures that the most suitable exemplars are determined and used for the clustering. The matrix computation is also iterative and updates after each iteration.
The affinity propagation algorithm for the LSP model is therefore summarized as follows:
1. Compute the similarity matrix between pairs of data points using the similarity matrix and the Euclidean distance metric
2. Activate the responsibility matrix (R), and compute R(i, k) the data points and identify exemplars.
3. Activate the availability matrix (A) and compute the availability of the data points A(i, k) for the data points. At this stage, data points consider the availability of other data points to be their centre points or exemplars.
4. Update the responsibility matrix R and availability matrix A respectively and iterate until when the matrices no longer change significantly, then a convergence is met.
5. Sum up the responsibility R and the availability A matrices for each data point
6. Identify exemplars with high responsibility as well as availability and associate them with the data points.
7. Assign each data point to the nearest exemplar based on similarity so as to form clusters.
8. At this stage, the states are fully clustered based on their LSPs
2.2.3 Classifier module
The classifier module makes use of the cluster means (CM) to classify the identified clusters and associated states into linguistic interpretations. The linguistic classification explains the mean rate or level of LSP for the states based on the clusters they find themselves. The linguistic classifiers are:
-
Lowest
-
Lower
-
Low
-
Average
-
High
-
Higher
-
Highest
The linguistic classifier "Lowest" represents the cluster having the least mean value while the classifier "Highest" is for clusters having the most mean value. The classifier module gives interpretation to the results obtained from the recommender module. It makes use of the linguistic variables to situate the clusters and states within them in relation to others. For instance, the classifier module will classify the states into respective clusters and determine the ones with highest/lowest rate of subscription. The results of the classifier module represents the recommender final output which the mobile network operators require for decision making.
2.2.4 The LSP Conceptual Model
Taking all these into consideration, the conceptual LSP model is thus presented;
In summary, the conceptual LSP model takes in a set of dataset from any of the mobile telecommunication providers across the states, the data extraction module computes the quarterly averages for Voice and Internet subscriptions. The Recommender module clusters the states based on their similarities using the affinity propagation. Finally, the classifier module makes use of the cluster means to classify the clusters into linguistic classification.
2.2.5 The Implementation Architecture of the LSP Model
The LSP model implementation architecture is also shown in Fig. 3 to enable easy adoption.
The LSP architecture comprises of subscribers who channel their calls through a specific mobile network provider. The subscribers are saved in a database which releases its contents to the LSP model on quarterly basis. The LSP model recommends to the mobile network owners on the suitable decisions that could improve their service delivery to the subscribers.