That chapter of his thesis focused on the framework SYSTEM-TH-001 A system designed to optimize the management of tourist services at a travel agency by using artificial intelligence and an architecture with bioinspired agents. SYSTEM-TH-001 It includes modules that update daily knowledge repositories of tourism providers and services. Although it requires manual intervention, the system automates and simplifies an administrator's tasks.
The framework was tested at the agency COMPANY-002 with a Multiagent System (MAS) based on an integrated and bioinspired organisation, using configurations that are adapted to the company's needs. Experiments have been conducted that evaluated key parameters and the satisfaction of customers or travel agents. The initial results indicated that customers prioritize attributes as price and quality of service.
The experiments compared the performance of SYSTEM-TH-001 with traditional methods and other algorithms, evaluating their ability to predict propensity for purchase. The bi-stage model of SYSTEM-TH-001 have shown a better balance between training and testing, avoiding problems with overlays while optimizing recommendations.
In actual tests, the system shown a significant improvement in conversion rate and reduced care time. It increased precise recommendations with benefits to both customers as well SYSTEM-TH-001 and an effective solution for digital transformation in the tourism sector.
3.1 Introduction
Today, tourism is one of the most profitable industries in Camilleri ( 2018 ). Each year, armies of tourists visit famous cities and historic vestiges or they simply have the pleasure to meet new cultures with other forms. Also, there are other tourists that simply seek calm and relaxation during their holidays with a great opportunity to enjoy the landscapes of Tucreac's dream beaches ( 2010 ).
All these human events result in the use of multiple transport types such as aircraft, trains and buses. ( 2018 ). In addition to generating millions of reservations at hotels, apartments and other similar properties that allow these tourists.
All this activity should be run by the travel agency, with information about destinations and preferences of different tourist types Moliner et al. ( 2007 ). That's why the emergence of focused systems to provide an end-user support (be it a tourist or travel agent) that will facilitate their booking process is completely necessary and currently at its disposal.
Also, travel agents should make an additional effort to provide their potential customers with the most interesting options in order To close them down. Due to this situation, it's natural that they start with the flowering of systems which will automatically make recommendations about travel and destinations by selecting best transport as required from Zapata et al. ( 2019 ).
All these systems have their limitations. The main shortnecks are the difficulty of drawing up budgets by travel agents. It's a task that requires great investment of time and effort, increased as the complexity with which to share destinations increases during your desired journey. Another important problem is the inability to provide feedback about services offered for customers. That problem was vital for travel agencies because they needed to know if the jobs that had been generated were suitable and customers have become satisfied with them after having enjoyed their contracted services. To this end, recommendations that filter initial customer preferences are key. They normally mark the evolution of tourism services offered as they tend to attract customer care from an early stage.
The COVID- 19 It was another big problem for the tourism sector because of restrictions on movements with a view to limiting virus expansion. That measure obviously had a serious negative impact on tourism, reducing travel and thus leading to high losses in the Bakar ( 2020 ). Nevertheless, progress with vaccine programmes and measures have been put at an end to these challenges.
The Commission's proposal for a directive lays down the following criteria:
For these reasons, a complete framework that can manage growing demand is crucial. That system should be capable of solving the identified weaknesses by best managing all problems that have been occurring since their client's interest in a tourist service, providing support to his travel agent as they select different possibilities and with an appropriate customer feedback after its end.
Thus, this thesis presents framewok SYSTEM-TH-001. That's an automatic system that includes multiple graphic assistants to provide complete support for travel agents and tourists. The novelties are as follows:
It covers all necessary processes with the process of selling tourist services, focusing mainly on customer interests and potential offers that best suit travel agency policy. The travel agent and client are guided by the system's recommendations at every stage of their sales process.
It generates budgets quickly in real time, leading to better decision-making at both an operational level and as travel agents with a strategic direction.
He's capable of learning through his experience and helps prioritize those customers or potential clients with a high propensity to buy. A model of propensity to purchase ML has been included in the system for this task.
The inclusion of an MAS and its intelligent agents best manage knowledge exchange with better communication. That MAS has been developed following the bioinspired abstraction of an antiguer, which manages work loads in Moreno et al. ( 2022 ), an important feature as most companies require real-time interactions with a view to meeting their customers' needs.
Provides support to the travel agent by helping them filter among best options depending on their client's interests. The system could be classified as a recommendation, which simplifies the processes to conduct at an agency's travel.
A number of experiments have been designed to show the value and best practice. The module dedicated to client prioritization has been evaluated accordingly.
independent so that we can check his added value. The system as a whole has been put to the test and very promising results have come about with this assessment.
3.2 Proposal
3.2.1 Automatic management of tourism
Automatic tourism management is one of the basic tasks within this area, due to a huge number and different processes for considering available offers as well with travel settings that have been put forward by Kler's customers ( 2009 ).
The management process relies on two approaches: the client and seller (for example, final resource or travel agency). The first is best developed by different studies and available systems. The problem relates to the implementation of new technologies and electronic turimso (for example, offers or marketing information about tourism services online). That means that a tourist service provider has direct contact with Hamid et al client 2021 ).
With regard to approaches that automate different tourism-related processes, most systems are system recommendations from both.
In the case of a travel agency approach, they typically work with clients who do not know where to ride or what their preferences are. That's why systems that are capable of collecting and processing information from customers can provide decision-making support or select the best provider with an agency for a particular Hsu et al client. ( 2013 ). The information can be contained from text Loh et al. ( 2003 ) or generated by creating Adriana profiles ( 2018 ).
On the client's approach, there are many challenges to work with. For example, systems capable of developing specific travel settings for clients with special needs are really useful Novikova et al. ( 2015 ). Also, there are other more generalist approaches and some examples of this have been those that focus on selecting the best path as referred to in weather time with traffic conditions Suframaniyaswamy et al. ( 2018 ), or those who recommend best trips for individuals and Herzog et al. ( 2018 ). They are all as virtual designers of Al Fararni and al travel plans. ( 2021 ) that are usually based on the customers' tastes and their trips have been made Garcia et al. ( 2009 ). For the selection of destinations, similar systems have been considered as
Two. They process the views of tourists to find their most popular destinations for them Zheng et al. ( 2018 ) and they also produce rankings of interest Alrasheed et al. ( 2020 ). Finally, the problem of offering optimal accommodation according to client's preferences has been also considered. These preferences are actually, the needs of customers Forouzandeh et al. ( 2021 ) and its economic capacity Ray et al. ( 2021 ).
The framework SYSTEM-TH-001 It's mainly focused on the travel agency perspective, providing specific elements for interacting with customers. In the first case, a salesman gets real-time recommendations consistent with information from customers about his or her interests. The system has a really new component to notify the seller about what customers have an increased tendency of buying their tourist service. That facilitates work and increases the agency's benefits. In the second case, customers can introduce their characteristics and interests. The system uses them to provide leaked information with the seller.
3.2.2 The framework SYSTEM-TH-001
The main goal of the framework SYSTEM-TH-001 That's to support customers and travel agents. The system therefore covers both perspectives related to the management of tourism services.
From a tourist perspective, SYSTEM-TH-001 provides different tourist services consistent with preferences and characteristics. For users of the travel agent model profile, this system offers best service with a view to customer-supplied preferences and previous performance ratings. Also, SYSTEM-TH-001 includes an extremely new functionality for a pyme, this is the ability to select those customers with more propensity about purchase at some time and that allows travel agents have their work plan priorities established as well while setting up necessary actions in each case.
The system is powered by a number of information sources that provide the necessary knowledge, such as: availability and price for flights or hotels.
The following sections explain the architecture of this framework SYSTEM-TH-001 considering different modules and focusing on the processes that cover their functionality.
Figure 3.1: architecture of the framework SYSTEM-TH-001.
3.2.3 General architecture and modules
The architecture of the framework SYSTEM-TH-001 It consists of five different modules: Information gatheer, Tourist service manager and Customer management. These modules are complemented with two knowledge databases that allow them to store this information: the Suppliers know how and Tourist services skills (see Fig. 3.1 ). These modules are capable of performing these three fundamental processes: selling tourist services, customer feedback and information update.
The Maintenance manager module serves as a central point for information maintenance and common parameters used by the other SYSTEM-TH-001. Typically some examples of information maintenance are related to changes in tourism resources (e.g. hotel name and new service providers or event closure). A mapping capacity of the same tourist service at different providers is necessary to facilitate system production with a range and prices. To them, use the Information Gatherer module to conduct these actions (see Fig. 3.2 ).
The Information gatheer module collects information from the virtual market of tourism. The market is based on information exchange between providers and consumers through the use of XML or JSON technologies. That module retrieves information of two types, static and dynamic. The first isn't often modified, as the name and description of these products are relevant to them. The second changes several times daily. Typically some information about this type of data are hotel rental or prices. All this information is stored in the Information Repository Suppliers knowledge. It is important to indicate that some of this information cannot be stored, because they are contractual with the providers and have only been consumed by
Figure 3.2: A vision of the maintenance manager and information gatheer module.
this module in a passive way.
The Repository Suppliers knowledge has the main goal to manage information acquired from providers. Note that this information is manually loaded into the repository by a human. The system uses all this information to make recommendations in an automated manner. The visual interface of SYSTEM-TH-001 provides several graphic assistants that certainly also help the user complete his tasks,
The Tourist service manager module is dedicated to risking potential client opportunities and generating custom offers for best recommendations (see Fig. 3.3 ). These recommendations are based on calculations made to get best tourist resources for an operation, ordered by interest. Each type of tourist resource is grouped by a category. The category of transport, flights and collective (e.g. trains or buses) as well with car rental services are included. In the category of accommodation, hotels and similar locations they are considered. Finally in the category of extras, excision and insurance they are classified or included.
The expression that provides best tourist resources to set up a complete tourism service (TS) is described as follows:
TS = max k) { 1 A) 124, E (12 and 3] TS k = X
C ∞ {T, A} θS ( 3.1 )
where T is transport, To housing and E are extras. The Commission's proposals have been put forward with a view to monitoring and evaluating the production of new technologies. 0 a 1 to indicate relevance. Note that the sum of pesos should be 1. S choreC is then the mean value of R and N values.
to select a service's tourist resources. S choreC j is defined as follows:
S choreC j = wEC u, g · f and e) 3.2 )
where j { 1,.).124 and C p845 have been referred to as Section 3 of this Regulation but are currently in force after 1 January 2012 with the following information: The values denoted by w are the weights of partial contributions to accordingly defined global score.
The element f C u, j are the preferred variables and expressed as follows:
N X
f C u, j = 1
g
i primerredC j, I ( 3.3 )
where you prefer C j, i is a value ∞ [ 0, 1 ] indicating feedback by customers on the basis of previous experiences. Also, f C or j is the mean value of customers' opinions.
For item f T and j, this depends on the categories concerned (N):
f T e, j = wt quality · qualityTJ + lt's stopopovers.topotesight JN and WL time 3.4 )
f A e, j = wa quality · qualityA. J.+ WA.location N. 3.5 )
f E e, j = WE quality · qualityY J + HE accessibility N flexibilitiand JM) services EN servicesA 3.6 )
In each category, a set of measures with values are considered [ 0, 1 ]. For the category Transport: quality measures transport performance, stops are number of potential stop and time will measure how soon we get to our destination. In the category Accommodation: quality measures how high an accommodation is, with its location expressed as relative to a tourist area and information about their level of service offered by them. Finally, for the category Extras: quality measures an extra and flexibility means that they have a capacity to reserve at times with their methods. Services expressed how much service had been provided while prices signified by its costs) It should be noted that the measures are weighted by w pesos with an amount of 1 for each category.
Through actions for each category, the system tries to address challenges that are key with a journey:
HOW: by defining the quality or level of a resource.
QUANDO: by grading the time necessary to reach your destination.
WHERE: by grading the place and its surroundings.
WHAT: by classifying acceptance of an extra resort as what it offers.
WHY: defines the set of extras and security offered by an appeal.
The module is responsible for filtering customers as they propose to complete their purchase. An algorithm that estimates the propensity to purchase will be charged with establishing this classification. That algorithm uses classical score processing techniques and two different ML methods in a Two-Layer architecture to categorize customers into five groups according with their booking probability. More details of this process will be submitted later.
The module includes three components: the tourist services organizer, our tour service filter and your custom feedback gatherer. The first includes a MAS to distribute the workload and get decision-making tasks shared. The second is in charge of filtering resources according to the peticioines made by customers and his third manages subsequent feedback.
In relation to MAS, it can carry out tasks related with ML methods using a bio- and antiguer-inspired structure that includes an old queen as well. The agent reigns the different tasks among his soldiers and builds up their final result. The soldiers organize their petitions and use them to solve the problem. That architecture of the intelligent agents is interested, because it allows to implement load balancing measures. Those measures consist of creating new and complete personnel, or ants to enhance the system's capacity. Further details about MAS will be provided in the following sections.
The Repository Tourist services knowledge stores information related to tourism service contracts for subsequent consultation and modification by travel agents. That information is subject to the management decisions and configurations made in The Tourist service manager module. It serves as a guide to the processes of composion or auto- generating tourist services. The feedback obtained from customers once the service has been completed is also considered.
Figure 3.3 Module vision: Tourist service manager, Travel agent managers and Customer management.
The Customer manager module has as its main functionality to support customers in performing the necessary tasks for creating real-time tourist services. That module communicates with the Tourist services knowledge and Suppliers know-how repositories through The Tourism service manager to conduct compositions of different elements that make up a tourist care. The service created is stored for subsequent verification and approval by a travel agent.
The Travel agent manager module aims at serving as a central internal management point for operations to create tourism services. It also operates a review of client-created services through the Customer manager module. It's directly connected to a graphic interface, enabling you with information about your customers' requests. These requests are used to consult with Tourist services knowledge and Suppliers Knowledge through the Tourism service manager module, thus obtaining interesting tourist care for users. The module can generate tourist services from scratch by selecting different options. Note that the main difference of this module with previous one lies in user's privilege configuration (travel agents have more privileges than customers).
The visual interface provides different options according to two perspectives: that of the travel agent and his or her client. For the first case, visualisation supports creating travel (for example a complete set of tourist services). A number of graphic assistants are trying to relieve the complexity and design, reducing client decision-making through a finite set. When the client approves how much travel service, they are sent to his or her Travel Agency for management: approval and modification. A number of exchanges have been held between the client and his or her travel agency until a firm confirmation by our customer
It's confirmed by the travel agency. Once the services have been confirmed and full payment verified, a sending process of all necessary documentation to enjoy them will be run. That documentation and its associated law are also made available to the client at his or her private user area for consultation with them. For the second perspective, that of a travel agent and similar functions are provided. Nevertheless, travel agents have an initiative here and can generate multiple tourist services for customers. More details about sales processes are described in the next sections.
3.2.4 algorithm for the detection of potential customers
The Tourist service manager module contains an estimate algorithm for propensity to purchase. It categorizes potential customers depending on their probability of closing a tourist service's reserve. That algorithm provides high-value information to the travel agency and allows Travel Agents To differentiate between good or bad customers (clients who have finally booked them). The algorithm consists of three fundamental elements: variable transformation, clustering and classification. The first element prepares inputs for probability estimation, performed by cluster and classification methods. The second item provides an initial probability of purchase and the last refines it at some cluster. The algorithm output categorizes potential customers according to five groups, commensurate with their conversion rate (for example the proportion of clients who finally reserve a tourist service).
The variable transformation element is performed by a Wight Of Evidence (WoE) Weed encoder 2005 ). That type of transformation and its benefits have been studied in the credit risk environment deeply Anderson ( 2007 ); Zdravevski et al. ( 2011 ). It helps to work with categorical and numerical data simultaneously, but can also run from missing information using an additional technique. Traditionally, WoE encoding is used to separate bad from good customers and that fits perfectly with an estimate algorithm for the propensity of purchase. The equation of the WoE transformation is:
WoE = ln% non _ events
% events ( 3.7 )
where the value of events refers to good customers and how best non- event represents bad ones.
One of the most interesting benefits to WIE's current approach is that variables with far too many discreet values are transformed into numerical and will be ordered according To a monotonic ratio withdependent variable
Figure 3.4: A biphasic model to estimate the probability of a propensity and reserve an airport service.
tit. WoE provides a numerical variable that regroup this large number of discreet values into an array and well populated categories expressing information for the new category. Another great advantage that can be achieved with the proposed solution is That WoE's a standardized value. The first allows using decision tree (DT) algorithms. Note that in a tree-based algorithm its prediction potential with categorical variables can be reduced using too many Friedman categories ( 2017 ). The second allows using distance-based algorithm as the k-means grouping algorithm, that requires variables with comparable values because it uses Mohamad and Usman eucliid ( 2013 ).
The algorithm is a biphasic model as and how shown in the figure 3.4 first applying grouping using k-means to get three clusters with different conversion rates (low, regular and high). Also to get more precise results a DT classifier acts on low and high clusters, for dividing them into lower but very bad. Thus, the proposed algorithm provides five production categories: very low and high. That type of output is calculated using the priori probability from propensity to purchase pons et al. ( 1997 ) and offers an interpretative result for a travel agent.
The model's key lies in how the variables are grouped into two phases. In the first phase, WIE transformation is applied to a set of variables that contain information about who's his client and where he comes from. Some examples of these variables are: contact channel (web, email or physical presence) and short- to long-term purchase if the journey relates with a particular event. With these variables, grouping is developed to first get three groups (low and regular) that will later be divided into the five categories
intended endings. Then in the second phase, WIE transponder's are applied to an inherent variable of his journey as number and place that will be visited by them. Note that WoE encoding should be applied independently to low and high probability groups, even with the same related variable as them because its goal is To capture patterns according for each independent group. Finally, two DT's applied to low and high probability clusters divide these groups into Two more sets each with five different categories of buy-offs.
3.2.5 Multiagent Organization (MAS) bio- inspired
Smart agents are independent entities that have been used by SYSTEM-TH-001 to manage ML methods. They set up co-operative activities organized through a MAS architecture.
The MAS propose consists of an ant that involves agents who play three different roles as ants: the queen, soldiers and workers (see Fig. 3.5 ). Thus, it is organized following a social structure of bioinspired ants. The queen (that's the managing agent) is responsible for an antiguer, receiving applications to analyze. Please note that several applications can be sent to this agent. The Realist agent then appoints the activity of evaluating elements from his application to a soldier officer (that is, an evaluator). That agent is responsible for allocating information to the workers. The latter process information and apply ML methods provided by the system. Note that there's a copy of the ML model available for every worker. Once the labor agents have completed their work, they join with them if necessary (protecting and monitoring an employee's or a group of workers) while returning his result to him.
That process relies on the model Belief-Desiri-Attempted (BDI) De Silva et al. 2020 ) where each agent has an aim or a set of them that have to be met in order To complete his and her life cycle. To this end, the queen has an assignedtak target and a soldier's goal to evaluate his information. All targets usually include associated tasks that are actions to be achieved. Smart agents have been put into place with at least one task, enabling them to achieve their goals. These tasks are put into place in an environment shared by agents, and this includes opportunities for potential customers. Each intelligent agent has a mental status and its processor or manager, as well with an array of beliefs (motivations). The Queen also has a number of rules and an opportunity-management plan, while at the same time soldiers have some similar standards to distribute knowledge about opportunities among workers. No em-
Figure 3.5: Extract from the main entities involved in Mas ants.
brego, workers include the ML model in their mental states to achieve information appraisal and a plan as well with some simple rules for organizing them. Finally, interactions between individuals follow the hierarchical structure that mainly consists of direct conversations. Note that in this case, workers do not have to engage with other employees as they address their commitment individually under orders from the soldier.
The design of the concrete model has been addressed through an agent methodology. SOFTWARE-TH-001. The result composition has been transformed to meet the MESA framework. The agent's conversations and interactions have been implemented following standards ORG-TH-006 ( ORG-TH-006 ) Melo et al. ( 2019 ).
3.2.6 Internal system processes
The work framework for SYSTEM-TH-001 presents three main processes in accordance with its architecture. The first process includes the different actions achieved by both travel agent and client until an actual sale of a tourist service. The second process refers to the task of feedback. Finally, the third process includes steps to update information about prices and services.
Figure 3.6: The offering and sale processes of a service for the provision with tourism services.
other information as appropriate.
The process of selling a tourist service includes an array and most have been common to both end users (that's, travel agent or client). It happens because both roles interacted several times during the process (see Fig. 3.6 ). Initially, end users login into the system. If users are already customers, they can apply for a specific tourist service and send them to the travel agency. However, if users are travel agents they can take the initiative (customers have been physically with them) or respond to potential requests from clients earlier. Then the system analyzes
Figure 3.7: The information update process based on client feedback.
customers to decide whether they are suitable for sale or not. If they are suitable, travel agents prioritize their demand. If customers responded to the response provided by travel agencies and one of their tourist services interested them, this system will complete its procedure for sending document with transaction information about tourism service. Nevertheless, if customers were unable to respond with the tourist services required by travel agencies. In this case, the proceedings are closed. Furthermore, customers may be unhappy with offers from the travel agency and will have to generate new services that meet their wishes.
The feedback process follows a number of steps with customers entering comments about an previously bought and completed tourist service (see Fig. 3.7 ). It's a simple and optional process performed only by customers where their opinions are stored in the system. The information they provide is used to optimize the recommendations contained in SYSTEM-TH-001.
The information update process is typical maintenance work of the system. It mainly modifies the knowledge stored in both repositories: Suppliers skills and Tourist services know how. That task should be performed daily to keep the framework up with date SYSTEM-TH-001. Nevertheless, while a manual process the system provides automated steps to manage actions and simplify work with an administrator. Please note that the administrator isn't considered an end user by your system (see Fig. 3.6 ),
3.3 Experiments
The majority of experiments, at least those most relevant to validation as a solution for performance are shown in this Section. SYSTEM-TH-001 3 experiments are included in actual environments. COMPANY-002 1 was the travel agency selected to conduct his experiments. Please note that this is the first company to have been put into production once development phase has come complete.
The MAS included in the system offers a basic configuration adapted to his or her needs. These needs have been previously tested by adjusting the parameters before these experiments. Thus, MAS consists of two ants with five soldiers and ten workers.
The first experiment covers the configuration of parameters and system setting. In addition, a basic test to evaluate the level of system satisfaction between travel agents and customers is submitted as required by recommendations. The second experiment details the performance of potential client detection algorithm. A comparison with other methods of art status is included. The third experiment evaluates system performance once put into place at the travel agency. Comparisons between travel agency benefits are addressed using and without the framework SYSTEM-TH-001. That fact illustrated the relevance of a system while managing an agency for travel.
3.3.1 Parameter setting and initial system validation
The first experiment includes a set of steps to show how the proposed system works. First, the framework parameters SYSTEM-TH-001 have been set up to produce recommendations about tourist services for customers and travel agents. These values were provided by 3 experts at the domain of a select travel agency (that's to say, people with responsibility for his company). Then, 10 customers (while preserving their anonymity) and 3 The company's travel agents (that is some of its workers) evaluated the recommendations obtained for their specific tourist wishes. Finally, these actors gave feedback about system performance and their satisfaction with the results obtained. That satisfaction has been measured by an Analytical Visual Scale of Satisfaction (MSA) Lynsky et al. 2021 ). That AVS was included in an anonymous document. The result of the AVS became a range from 0 a 1 to get the resultant opinion.
1 [LINK-REDACTED]
Rol Satis VAS Agent A 0,85 Agent B 0,93 Agent C 0,73 Mean ± S.D. 0.84 ± 0.10
Table 3.1: Value of satisfaction provided by travel agents.
Rol Satis VAS Customer 1 0,71 Customer 2 0,62 Customer 3 0,89 Customer 4 0,76 Customer 5 0,95 Customer 6 0,58 Customer 7 0,78 Customer 8 0,69 Customer 9 0,92 Customer 10 0,76 mean ± D.E. 0.77 ± 0.12
Table 3.2: Value of satisfaction provided by clients.
The parameters set up (see Section) 3.2.3 ), the travel agency's experts decided as follows values for them. By θT. ÇA and ΔE, values have been set at 0,3, 0,45 and 0,25 respectively. That decision was motivated by the assumption that customers give more importance to accommodation and location than other tourist service features. And then, wC u and j y.g. 0,4 and 0,6 giving moderate relevance to comments from previous customers. The weights of the elements in each category (transport, accommodation and extras) are established. For transport category, wT quality and Wt stopovers are set up at 0,3, 0,1, 0,2 and 0,4 respectively. That decision was motivated by the assumption that users often consider transport prices to be very important above their other characteristic. For the category of accommodation, WA quality and location have been set up at 0,35, 0,25, 0,15 and 0,25 respectively. These values reflect an assumption that users often consider as relevant the price and quality of their hostage, unless its location or services are also very important. For the category of extras, WE quality and WE flexibility. 0,45, 0,15, 0,15 and 0,25 respectively. That decision starts with the premise that users place more emphasis on prices and quality of extras.
Once system parameters are fully set, the three travel agents make requests to SYSTEM-TH-001 To get recommendations according to your preferences. These preferences were selected as heterogeneous to validate the system through a broad range of possibilities.
The customers then provided indications that the system was currently being assisted by travel agents to solve potential problems with how they operated. Travel agencies also provided users with best system-obtained settings. If a client agreed with one of the tourist services offered, his or her system completed its corresponding booking process.
Finally, an anonymous document with the AVS was submitted to participants. The results are shown in the table 3.1 and in the table 3.2. The average satisfaction of travel agents was 0,84 while the average number of customers was 0,77. In both cases, similar standard deviations were obtained. Thus, the system's overall average satisfaction was 0, 80 ± 0, 11.
To conclude, we can say that the result is quite acceptable from both perspectives (customers and travel agents). Also, customers are more pressing and demanding than travel agents in evaluating the satisfaction of recommendations made by their work framework SYSTEM-TH-001.
3.3.2 Validation of the estimation algorithm from propensity to purchase.
That experiment addresses the performance of an estimate model to propose purchase (bi-stage). The proposal compares with the most typical Sen et al. ( 2020 ); González et al. ( 2020 ): Random Forest (RF and Gradient Boosting, K-Nearest Neighbbour.
COMPANY-002 have provided the dataset for development of this experiment. Contains about 104 with an initial conversion rate of 14 % That is, the percentage of customers who have finally bought their tourist service (that's to say data are unbalanced).Each instance includes a variety and diversity with information about contact dataset. In this particular case, more precise information about the data are available as a result of confidentiality. All selected algorithms are trained with 80 % and are tested with 20 % remains. That division is performed with stratified sampling. The hyperparameters of ML alternatives are select by 10 -fold.cross.- validation. The output of each algorithm is presented in five categories that care for probability distribution
RF GB KNN LR Bi-stage
model freq. c.rate frq.tit
Train
low 75.0 0.0 74.2 5.8 0.0 100.0 72.5 11.5 16.6 1.7 low 9.9 0.0 10.3 25.2 54.2 0.0 9.0 17.8 17.1 5.1 regular 9.9 89.1 10.3 34.5 29.8 20.6 13.2 19.6 44.7 10.9 high 3.9 100.0 4.1 63.4 11.9 42.2 4.2 26.7 15.8 32.8 very high 1.2 100.0 1.0 87.3 4.1 100.0 1.1 28.2 5.9 48.0
Test
low 57.5 4.9 73.6 7.0 0.0 - 72.3 6.7 17.0 2.5 low 17.1 16.5 10.0 25.0 75.2 7.5 2.0 25.0 17.0 6.7 regular 25.2 30.9 9.8 28.8 17.4 26.7 5.8 22.9 46.1 10.2 high 0.0 - 4.8 41.4 5.3 40.6 7.3 34.1 14.6 34.2 very high 0.2 100.0 1.8 63.6 2.0 58.3 12.6 35.5 5.4 51.5
Table 3.3: Performance of ML classification algorithms with relative frequency (frq.) expressed as a percentage about the general total and conversion rate for each category
from the training set. Performance is measured in terms of conversion rate and relative frequency for each classification group expressed as a percentage from the general total. The predictive power is evaluated by checking the conversion rate in different groups and frequency provides a group's significance. Finally, the training's potential overlays are evaluated and conformed with a test console performed.
The results are shown in the table 3.3. RF has a bad match between training and testing both at frequencies as with conversion rates. In fact, one group does not have representation in the dataset of test. KNN has a similar conduct at training and testing sets with differences from 21 %. In this situation there's no point in evaluating any other performance measures. GB has a generally good matching between training and testing instances, with the exception of conversion rates for high or very low categories that have differences more than 20 %. These differences mean, as with RF and KNN) that it makes no sense to evaluate the other items of assessment. LR has the best concurrently between training and testing, both in frequency as with conversion rates. The distribution of data is high and unbalanced in all categories, where the low probability group includes more than 70 % information. Moreover, the five categories do not have sufficient distance between them at an end of their conversion rate. 2 % In some cases, this makes differentiation between close groups difficult. Eventually, the proposed bi- stage model has an excellent matching between training and testing instances with a maximum deviation from 3 %. The data distribution is the most balanced of proven models, where regular category with 45 % data and the very high at least dense with 6 % information for training and 46 % and 5 % for the test set respectively. The probability categories obtained are definitely different with five likelihood values
Traditional Innovators Services n Time c.rate Re.g.
Web
N Coasts 20 3.2 15.0 2 20 2.9 25.0 3 N Islands 20 5.2 10.0 2 20 4.8 15.0 3 Dubai 16 10.1 6.3 1 16 9.6 12.5 2 Safaries 12 11.2 16.7 1 12 10.7 16.7 2
Other
N Coasts 8 4.2 37.5 1 8 2.7 50.0 2 N Islands 8 5.7 25.0 0 8 3.0 37.5 1 Dubai 4 12.0 25.0 1 4 8.0 50.0 1 Safaries 4 12.5 25.0 0 4 9.0 25.0 1
Table 3.4: Abstract of the results obtained for traditional and innovators groups with respect to each tourist service. N's are how many potential customers have been found in every care or group The time is the average (in hours) that each travel agent passes by processing his or her tourist service until it's accepted and ruled out. The conversion rate and number of recommendations (Rec.) are presented
and is valid in the test dataset with minimum deviations.
The experiment conducted shows that this particular problem needs an ad-hoc approach, such as the bi- stage model put forward. That solution was appropriate to solve the problem of propensity for purchase provided by COMPANY-002. The bi-stage model best classifies in different probability categories with balanced frequencies. It also provides the best match between training and testing, thus ensuring an absence of overlays. Thus, according to performance requirements the proposed method better serves as a last-generation classifier for today's target.
3.3.3 Evaluation of complete system efficiency
In this experiment we can check the efficiency of SYSTEM-TH-001 in a real situation. The tuning of the system with parameter configuration has been maintained and they have remained at first stage.
The experiment starts by organizing randomly and proportionally to potential clients of 184 in two groups: Traditional and Innovators, to make comparisons between them. They all seek the same type of tourist services: National Coast, national lands and Dubai-Maldives. Also, web and other channels are referred to as phone calls with family recommendations or face-care visits at the travel agency. It should be noted that these services cover a range of dedicated tourist care.
The traditional group consists of 92 customers that have been cared for while maintaining the traditional way in which our company was used (i.e. without framework support)
SYSTEM-TH-001 ). On the other hand, innovators have been formed by 92 customers that were cared for using the frame SYSTEM-TH-001 to carry out users' requests and the settings provided by a travel agency. Both groups worked with the same travel agency to minimize potential changes in results.
The experiment considers three main variables: the mean time (in hours) that each agent dedicated to processing his or her tourist service offered with customers, conversion rate and number of recommendations. These results are presented in the table 3.4 for both groups.
For the traditional group, results show that an average time to manage services is lower at national destinations. That means, customers are more demanding when the destination of their journey isn't in that same country. The conversion rate fluctuates on the web channel with low value for Dubai-Maldives service (that is, only a few customers complete their purchase) while Safaris and National Coasts have high percentage. A total of recommendations have been made 8 for this group.
For the innovators group, Dubai-Maldives has its lowest conversion rate and National Coasts have his highest at web channels. Significant differences have been found in the number of hours spent, and with regard to recommendations as well at a conversion rate between traditional or innovators. The system had been an acceptable increase in the conversion rate, a significant reduction of hours and more recommendations. A total of 15 of recommendations for the Innovators group. The changes in conversion rate and number of hours dedicated are motivated by the propensity to purchase estimate algorithm. He allowed to select best customers for the travel agency, and his clients with less interest in buying them have been ruled out. Moreover, visual information provided by the system and precise recommendations could have also been easier for customers to accept their tourist service.
SYSTEM-TH-001 has shown that it works perfectly with the daily problems of a travel company. It could also increase the conversion rate that translates into benefits for our company and reduce their number of hours until customers accept or discard different opportunities available to a particular tourist service. That fact translates into less stressful situations for employees and more agility in the booking process.
3.4 Conclusions
The chapter about managing tourism services through client scouting using a bioinspired agent architecture concludes that the framework SYSTEM-TH-001 It represents an innovative and efficient solution for digital transformation in the tourism sector. The system combines advanced artificial intelligence tools and a bioinspired approach to optimize interplay between travel agents with customers. Through its modular architecture, SYSTEM-TH-001 facilitates the collection, processing and updating of relevant information by improving both efficiency as well how precise recommendations are.
The analysis of experiments performed at COMPANY-002 shown that SYSTEM-TH-001 outweighs traditional methods and standard algorithms in terms of conversion rate, customer satisfaction or shorter care times. The bi-stage model, integrated into the framework and emphasized its ability to predict propensity for purchase with high precision enabling agents To prioritize customers best potential.
The implementation of a bioinspired multiagent system added flexibility and scalability, ensuring an efficient load balance with real-time response capacity. Moreover, the dual approach to SYSTEM-TH-001 That benefits both agents and customers, strengthens their potential as a key tool to address today's challenges in the tourism sector with its status of an innovation standard.