Strategy Mining for Inferring Business Information System User Intentions

Abstract

The aim of this study was to identify user strategies to infer their intentions in developing activities, from the current process model and the real process model of the business. A user’s intentions can be used to identify their behavior, and to define the requirements for improving the business information system. The presented method follows the guidelines of the current mining tools, and it is supported by a knowledge base of businesses in general and the event log for a specific business. The user strategies are validated and weighted through the rules of the specific business. The user intentions are inferred based on their strategies and the knowledge of an expert within the specific business. The method is applied to a specific sales business, and the obtained results suggest that the proposed method can extract 75% of user intentions. In addition, the method is generalized to apply it to any business, as long as we can obtain the event log and the rules of the business.

1. Introduction

In the context of a business where both the current process [1] (written in the business functions and procedures manual) and the user profiles of the information system have been defined, business activities are not always developed by observing these precepts [2]; this can be evidenced in the content of the event log, in contrast to the current process content, which implies the existence of the business’s real process [3,4].

Current business processes do not have the flexibility to adapt to the demands of usability and execution opportunities, according to the demands of an increasingly dynamic environment, because static models (e.g., he current process model), which are exclusively oriented towards the prescribed activity, govern these processes [5]. The users of information systems, each time, try to develop their activities in the shortest possible time; to this aim, they use their own strategies, which allow them to achieve their objectives, according to their profiles [6]. Therefore, these users pretend to develop their daily activities, with the intention of improving their performance, without affecting the business objectives. The modern information systems, which are perfectly aligned with the enterprise objectives, and without losing sight of the business rules, generate the event log that contains the activities according to the current process; in addition, this event log contains the user strategies and the activities and strategies of the real process [5,7].

The present work aims to identify and formalize the intentions of the user of a business information system based on their strategies. In Figure 1, we present a summary of the method that we developed.

The method was developed for the business information system context and is centered on the user strategies. In summary, we developed the following: a knowledge base for any business in general, an event log for a specific business, a method to perform the mining of these resources, the user strategies obtained as a result of the mining applied to the sale-specific business, and the user intention inference of the business information system. For the development of this novel approach, some relevant concepts are specified below.

Business in general: We considered the business activities of all businesses, and we structured a knowledge base (KB) [8] containing activities of all kinds. This KB allowed us to validate the activities that take place in a specific business; in our case, we selected a sales business (any other business can be selected).

Specific business: We considered a formal business, where the daily activities are developed through an information system and some databases. As a result of the activity processing in the business, a generic event log (GEL) [9,10] is produced, which stores the business’s activities.

Users: Business information system users develop the daily activities in the specific business; this is carried out through the information system [11] using the business’s current process and their own strategies.

Business rules: Formal business statements that allow businesses to reach their aims through activities are developed [11]. Based on the business rules, the heuristic rules are deduced to validate the user activities.

User strategies: Initially, new users develop their activities according to the current processes [12], but as users gain experience, they try to perform the same activities to save resources, specifically, the execution time of daily tasks, which is achieved using their own strategies.

Event log: Flat file containing the records of events (process activities), recorded by information systems, as a product of the development of a business’s daily activities. An event log [9] is a collection of timestamped event records produced by the execution of a business process. Each event record tells us something about the execution of a work item (a task) of the process.

Quality event log (QEL): It is GEL corrected and debugged for a specific purpose and to improve the quality of process mining and intention mining [10]. The GEL is an event log that is generated by default via the information system in any business [10].

Intention mining: User strategies (user intentions) are determined through process mining techniques and tools [7] applied to the quality event log.

User intentions: Without losing sight of the business’s rules, and in accordance with the maturity and experience acquired, a user performs their daily activities with the intentions of reaching their goals according to the business objectives, trying to save resources and improve the quality of the results. In the execution of a certain process instance, the user develops their activities; additionally, they use their own strategies (intentions) to fulfill their goals more efficiently. Each user intention can be fulfilled with several strategies, and each strategy can be used to fulfill several intentions [5].

Current processes (e.g., business process manual) are promulgated processes that are used to develop activities and reach business objectives [10]. Real processes are the processes that are actually executed within a business (combination of current processes and user strategies) [10].

The present work, in addition to this section, is described in Section 2, which reviews related works; in Section 3, the materials used are described; in Section 4, the proposed method is specified in detail; Section 5 presents the experimental results; Section 6 presents a discussion of the work; and in Section 7, the conclusions obtained from the work and possible future work are presented. Finally, the references that support this work are listed.

2. Related Works

In

Table 1. Comparison of the present work with previous works.

Previous Author	Aim	Data and Tools	Result
Present Work	Discovery the user intentions of business information system	NewsIR’16 repository Knowledge base for any business Event log for specific business Business rules Knowledge expert	User intentions, possibility of user behavior inference, and improving information system
Epure [12]	Automate the construction of intentional process models	Event log Hidden Markov models	Map Miner Method
Khodabandelou [5,7]	Toward the construction of process model topology	Eclipse UDC processes Map Miner Method	Intentional process models
Deneckere [13]	Design a technique to mine intentional models from traces of non-deterministic activities that follow a stochastic process	Activity traces Supervised machine learning	Intentional process models
Huang [14]	Classifying intentions by category	Linguistic patterns Issue reports	Intention categories Recommendations for software developers
Cohan-Sujay [15]	Analysis of intentional business functions that can be performed effectively on short texts	Short texts (simple sentences) Machine learning	Words that express intentions
De Sa [8]	Knowledge base construction	Deep Dive (databases and machine learning)	Knowledge base

, we present a brief comparison between the aim, data, tools, and results of the present work and the previous works.

Epure [12] developed a tool to build intentional process models; in our work, we extracted intentions from these process models.

Khodabandelou [5,7], based on the work of Epure [12], generated intentional process models; in our work, we formalized these intentions to infer the user behavior.

Deneckere [13] and Khodabandelou [5,7] generated the same intentional process models, but using different resources; in our case, we worked from the non-structured texts (news) and a structured log.

Huang [14] and Cohan–Sujay [15] obtained their patterns (words and sentences) based on linguistic patterns and machine learning; in our case, we used NLP functions.

De Sa [8] constructed a knowledge base with the DeepDive function; in our case, we used our own method.

In addition, in [16], a previous work of the present authors, the most important advances in intention mining can be reviewed.

2.1. Relevant Concepts

Process mining techniques are included among the data mining and machine learning techniques [7]. Modeling user behaviors in terms of activities and ignoring the underlying human cognitive operators, such as intentions and strategies, are the goals of process mining techniques [5]. User strategies can become hidden processes [17], and process mining tools allow these processes to be represented in terms of activities [13,17].

Many researchers in the field of intentional process modeling have demonstrated that the fundamental nature of processes is mostly intentional; therefore, the processes should be modeled from an intentional point of view. In this regard, nowadays, intentional process models have emerged to offer a flexible structure to model processes [5].

In some works proposed in the literature, the main objective of the use of intention mining is to extract sequences of user activities from the event logs to evaluate and predict the intentions of users regarding those activities, as reviewed in [18]. Process models focus only on activities, and intentional process models focus on the intentions underlying the activities, rather than the activities themselves [19].

A set of strategies allows users to achieve their intentions, and a strategy can be used to achieve several intentions [14]. The relation between intentions, strategies, and activities represents the top-down structure of reasoning and acting in the cognitive processes of the human brain [5]. On the other hand, a sub-intention is associated with a parent’s intention, and one intention is fulfilled if at least one of its children’s sub-intentions is fulfilled [20].

Several support activities and guidance solutions based on process mining have been proposed, but they lack suitable semantics for human reasoning and decision making, and mainly rely on low-level activities [13]. Process mining aims to discover, verify the conformance of, and enhance the activity-oriented process models from the event log. Intention mining has the same objectives as process mining, but it specifically addresses intentional process models [21], i.e., the process’s focus on the reasoning behind the activities.

2.2. Intention Mining Applications

The Map Miner Method (MMM) is applied to a real-world dataset, which is the event log of Eclipse UDC (Usage Data Collector) developers [19,20]. The resulting map process model provides a valuable understanding of the processes followed by developers and provides feedback on the effectiveness and the demonstrated scalability of the MMM.

A map specifying intentions and strategies for entity/relationship modeling was given to the students as guidance [13,19]. To obtain traces, they developed a web-based tool, which records which sections of the map were followed by the students while creating an entity–relationship diagram.

To demonstrate the validity of the FlexPAISSeer [12] (Flex Process-Aware Information System) tool, a single case study was conducted. The case company was selected considering its suitability (the support of flexible processes through its software product): the childcare system developed via 42 windmills used by several child daycare centers in the Netherlands.

Finally, the ProM framework is a pluggable framework that supports various plugins for different process mining techniques, such as the α-algorithm and its extensions [13].

2.3. Suggested Future Works by Previous Authors

Intentional process mining might help improve guidance, provide better recommendations, facilitate process modeling and process model quality assessment, identify the gap between prescribed business requirements and goals, and help CEOs assess and monitor strategic goal implementation [13].

To establish a base to manually infer the names of strategies and intentions, intentional process mining can be fully automated by building sophisticated ontologies from the uncovered topics. These ontologies should consider the context in which the processes are enacted, as well as current situations [20].

A ProM plugin should be developed for IntentMiner and IntentRecommender. Additionally, official XES extensions [5] should be proposed to integrate the concepts of context and entity to the process.

3. Materials

In this work, several software tools, patterns, templates, data, repositories, and other resources necessary for the implementation of the proposed techniques were used. The proposed algorithm was applied to Ubuntu, due to its security and free distribution advantages, few system requirements, and ease of use. To store and manage the KB, the PostgreSQL database manager and pgAdmin III were chosen. For the coding, the Python programming language was used. In addition, it was vital to include other ancillary tools and data sets, such as Stanford’s CoreNLP natural language processing (NLP) system, with the Penn Treebank tag set, as well as the Signal Media One Million (NewsIR’16) [22] news article dataset, which was downloaded from the Kaggle repository.

4. Method

The proposed method was based on a KB where the information of any business, in general, is stored [8], along with a record of events [13] of the specific business (sales), which corresponds to the activities of the sales business and the user strategies. In addition, a set of heuristic rules was established from the specific business rule, which helped to validate the possible strategies of the users. The user intentions were determined from the user strategies and the judgment of a commercial sales expert. This method consists of the stages shown in Figure 2 and in the GitHub repository [23]; the Python code and the tables used in the development of this research are stored and made public.

4.1. Building the Knowledge Base

To determine, understand, and conceptualize the activities of a business, we created a KB in accordance with the policies and rules of the business proposed in [8]. The following describes the creation of our KB step by step.

4.1.1. Knowledge Base Structure

The structure of the table set with their attributes and the pertinent data types were created in the Postgres DBMS. The data tables and their dictionaries were stored and published in the GitHub repository [23].

4.1.2. Data Preparation and Load

Article Extraction

The text was extracted from 200 documents published in open access multidisciplinary business repositories. Moreover, this information was loaded into the article data table (articles = sample extract (NewsIR’16 [22])).

Sentences Extraction

The textual content of each document was tokenized, i.e., the text was divided into sentences, and every sentence was divided into words (tokens). Tokenized and lemmatized sentences were stored in the sentence data table.

sentences = lemmatized (tokenized(parsing(document text))).

The text was divided into sentences, the sentences were tokenized, the tokens were lemmatized, and the tokenized and lemmatized sentences were stored in the sentence data table.

NLP Tag Assignment (POS Tag and NER Tag)

Each token in every sentence was assigned a labeled part of speech (POS) using Stanford’s CoreNLP system [24] and the standard “Penn Treebank tagset”.

We used Stanford’s CoreNLP system in the tag assignment, named entity recognition (NER), according to

Table 2. Tokens and named entity recognition.

Token Tag	Named Entity Recognition	Meaning
VB, VBD, VBG, VBN, VBP, VBZ	ACTIVITY	Verbs in all modes and times
NN, NNS	NOUN	Common nouns
NNP, NNPS	PERSON/ORGANIZATION	Nouns
JJ	ADJECTIVE	Adjectives
JJR	COMPARATIVE ADJECTIVE	Comparative adjectives
JJS	SUPERLATIVE ADJECTIVES	Superlative adjectives
RB	ADVERB	Adverbs
RBR	ADVERB ADJECTIVE	Comparative adverbs
RBS	ADVERB ADJECTIVES	Superlative adverbs

. Each token was assigned a corresponding POS tag; each token, according to their POS tag, was assigned an NER tag; the sentences tagged with POS and NER were updated.

Strategy Mentions Extraction

From the sentence data table, we extracted the tagged tokens that had sequences corresponding to the generic structures of sentences following English grammar rules.

The sequence tokens (NER-tagged) of each sentence were extracted according to the following structures: “VERB”, or “NOUN” + “VERB”, or “NOUN” + “VERB” + “NOUN”, or “ADJECTIVE” + “NOUN” + “VERB”, or “NOUN” + “VERB” + “ADVERB”, or “VERB” + “ADVERB”, or “ADVERB” + “VERB”; each sequence token that was extracted corresponded to an activity (strategy); each NER-tagged token of a strategy was replaced by their token (word); and the structured strategies were stored in the strategy_mention data table.

Candidate Strategies Extraction

We generated the candidate_strategy data table based on the strategies stored in the strategy_mention data table.

For each strategy, the sentence structure was verified, the duplicate strategies were eliminated, the punctuation marks were eliminated, the nonprintable characters were eliminated, and the debugged strategies were stored in the candidate_strategy data table. In

Table 3. Candidate strategies.

Strategy Name
10-min
22-year-old
39-year-old
6-foot-2
ability
ability to accept
accept
accomplish goal
American news sales
approach
business show
business performance tdameritrade.com
busy
buy
economically
economy
efficient business show
forward-looking information involved
forward-looking statement is
forward-looking statement includes
negotiation for effective preparation
problem-solving negotiation
problem-solving techniques
quota set
sale
sale say
sell
seller amount
sensitive information included
technologically improve system
traditional custom
weakness knows best
work collaboratively

is shown an extract of this data table, and the set of strategies (S) of KB (all business in general) were defined as below.

S = {∀x/x ∈ KB and (x = ”VERB” or x = “NOUN”+“VERB” or x = “NOUN”+“VERB”+“NOUN” or x = “ADJECTIVE” + “NOUN”+“VERB” or x = “NOUN”+“VERB” + “ADVERB” or x = ”VERB” + ”ADVERB” or x = ”ADVERB” + ”VERB”)}

4.2. Quality Event Log Generation

From the GEL, which was obtained from the information system of the specific business (sales), the QEL was generated through the process of debugging [9,10]; this QEL contained the sales business’ strategies and the user strategies. The registers of QEL were stored in the user_strategy data table of KB and are presented in

Table 4. Quality event log (QEL).

Trace ID	Timestamp	Activity
1	2019-11-20:10.20.30	Customer service
1	2019-11-20:15.20.31	Generating customer order
1	2019-11-20:23.10.34	Canceling insufficient stock order
2	2019-11-20:11.10.35	Customer service
3	2019-11-20:11.10.35	Customer service
3	2019-11-20:11.10.36	Generating customer order
3	2019-11-20:11.10.37	Local stock control
3	2019-11-20:11.10.38	Delivering customer order
3	2019-11-20:11.10.39	Registering sale
3	2019-11-20:11.10.40	Sales records
4	2019-11-20:10.21.41	Customer service
4	2019-11-20:15.21.42	Generating customer order
4	2019-11-20:16.21.43	Local stock control
4	2019-11-20:22.21.44	Delivering customer order
4	2019-11-21:08.21.46	Registering sale
4	2019-11-21:23.21.45	Ordering home delivery
5	2019-11-20:15.30.31	Customer service
5	2019-11-20:18.30.32	Generating customer order
5	2019-11-20:19.30.33	Local stock control
5	2019-11-20:20.30.34	Delivering customer order
5	2019-11-21:10.30.37	Billing custom sale
7	2019-11-21:14.30.43	Customer service
7	2019-11-21:14.35.44	Generating customer order
7	2019-11-21:14.50.45	Local stock control
7	2019-11-21:16.30.46	Completing customer order
7	2019-11-21:16.45.47	Remote stock control
7	2019-11-21:17.30.48	Producing customer quotation
8	2019-11-21:14.30.49	Customer service
8	2019-11-21:15.30.50	Generating customer order
8	2019-11-21:17.30.51	Local stock control
8	2019-11-21:18.30.52	Completing customer order
8	2019-11-21:19.30.53	Remote stock control
8	2019-11-21:17.50.54	Producing customer quotation
8	2019-11-21:17.55.55	Canceling customer order
9	2019-11-21:16.55.56	Customer service
9	2019-11-21:17.55.57	Generating customer order
9	2019-11-21:19.55.58	Local stock control
9	2019-11-21:23.15.02	Completing customer order
9	2019-11-21:23.35.31	Remote stock control
9	2019-11-22:09.55.42	Producing customer quotation
9	2019-11-22:11.55.25	Delivering customer order
9	2019-11-22:11.55.08	Registering sale
9	2019-11-22:12.55.33	Sales records
9	2019-11-22:20.55.12	Ordering home delivery

.

The QEL contained the strategies of the real process (SRP) of the specific business, according to the real process model (Figure 3).

SRP = {∀x/x ∈ QEL}

(1)

Then, the strategies of the specific business (SSB) were defined as below.

SSB = S ∩ SRP

(2)

4.3. Strategy Mining Method

The KB created above contains the data and information for any business (in general). In our case, we considered a sales business (in particular); the method was designed using the steps specified below.

4.3.1. Business Rules Specification

The KB that was structured and fed with data of a general nature contains activities to satisfy the rules of any business; examples of the rules for any business can be seen in [10]. Nevertheless, in the case of a sales business [21], we defined the following rules: to order or quote, the client must be registered; the order or quote can be completed with local or remote stock; an unconfirmed order or quote compromises the stock for a certain time; in accordance with the business’s policies, a confirmed order or quote is registered as a sale and affects the local and/or remote stock. A sale can be sent home, and each seller (user) can use their own strategies to improve their performance while observing the business’s policies and rules.

4.3.2. Identification of Business’s Current Strategies (BCS)

We specify the sales business’s current strategies below [9]. The activities model for this business can be seen in Figure 4, which consists of: customer service, generating customer order, producing customer quotations, completing customer order, local stock control, remote stock control, delivering customer order, canceling customer order, registering sale, ordering home delivery, and sales records.

BCS = {Current strategies catalog of the specific business}

(3)

4.3.3. Definition of the Model of Business’s Current Activities

We generated a sales business activities model [13,17] according to the sales business’s current activities, using the alpha miner algorithm of ProM. This model is shown in Figure 2.

4.3.4. Definition of the Model of Business’s Real Activities

Based on the activities of the QEL, which contains the real activities, the real activities model of the sales business is presented in Figure 3.

4.3.5. Definition of the Business’s Heuristic Rules (BHRs)

Each business has its own rules, under which their activities develop. In the specific case of the sales business, the heuristic rules are defined below, which are based on the sales business’s rules [11,25,26]:

The strategy is a sentence that has one of the following structures.

 

“VERB”.

“NOUN VERB” or “VERB NOUN” or “VERB ADVERB” or “ADVERB VERB”

“NOUN VERB NOUN” or “ADJECTIVE NOUN VERB” or “NOUN VERB ADVERB” or “NOUN VERB ADVERB” or “NOUN ADVERB VERB”.

The strategies contain words related to the business’s domain (sales).

 

DEALING (“order”, “quotation”, “stock”, “sale”, “price”, “customer”, and “user”).

TRADING (“sale”, “sell”, “seller”, “purchase”, “buy”, “buyer”, “bill”, and “billing”)

BHRs = {Heuristic rules catalog of the specific business}

(4)

4.3.6. User Strategy (US) Extraction

The difference between the real process and the current process (Figure 3 and Figure 4) is the user strategies. Therefore,

US = SSB − BCS

(5)

4.3.7. User Strategies—Validated and Weighted (USW)

The user strategies were weighted according to the heuristic rules, which are specified on the BHRs (5). For this, the model is presented below.

W:

weight of BHRs.

w:

BHR compliance weighting by the US.

n =

count (BHRs): (“n”: number of business’s heuristic rules).

$$\mathrm{m}={{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{W}}_{\mathrm{i}}\right)$$

$$\mathrm{USW}={{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{w}}_{\mathrm{i}}\right)/\mathrm{m}$$

(6)

4.4. User Intentions Inference

Finally, the steps to formalize user strategies are specified below.

Weighted user strategies were stored in the strategy_weight data table: from the strategy_weight data table, the more relevant (those which have a weight > 0) user strategies were extracted; user strategies with a weight > 0 were stored in the user_strategy data table, as listed in

Table 5. Weighted user strategies.

Strategy Name	Weighting
billing custom sale	0.75
ordering home delivery	0.5
sale register	0.33
sale say	0.33
American news sales	0.25
pay seller amount	0.25
couple sales	0.17
news sales	0.17
sale	0.17
seller amount	0.17
buy	0.08
sell	0.08

; and based on this table, we can infer user intentions from the participation of the specific business’s sales expert [27,28,29,30], as shown in

Table 6. User intentions.

Strategy	Intention
billing custom sale	Win over the customer
ordering home delivery	Customer satisfaction
sale register	Control the business
sale say	Advertising, promotion
American news sales	Listen to the competition cautiously
pay seller amount	Timely payments
couple sales	Complement yourself with a partner
news sales
sale
seller amount	Awards, incentives
buy
sell

.

5. Results

The user strategies were identified from the differences between the current process and the real process of the business (Figure 3 and Figure 4). Based on the user strategies, we inferred their intentions in the development of business activities. The experimental results are presented below.

5.1. Business Knowledge Base

5501 sentences in general.

Tokenized and lemmatized sentences.

2458 activity sentences (strategies).

5.2. Quality Event Log

5004 sales business activities.

Sales business activities verified by KB.

Sales business activities validated by BHU.

44 user strategies.

5.3. User strategies—Validated and Weighted

12 validated and weighed user strategies.

Table 5 contains the weighted user strategies.

75% maximum weighting.

8% minimum weighting.

5.4. Inferred User Intentions

Table 6 contrains user intentions.

8 user intentions.

In summary:

 

Resources:

 

200 articles of the NewsIR’16 repository [22];

5004 sentences of the GEL (sales business event log).

Results:

 

5501 sentences in general;

2458 activity mentions in KB (sentences that denote an activity);

44 user strategies verified in KB;

12 user strategies validated and weighted by BHRs;

8 user intentions.

6. Discussion

In the KB construction, the use of Python and NLP was very useful for parsing (obtain sentences), tokenizing (obtain tokens/words), and lemmatizing (determine meaning of words).

In this method, two practical mining strategies have been developed: first, sentence mining, by the extraction of sentences from news articles and the structuring of the KB; second, the mining of user strategies from sentences of KB and QEL, by the extraction of the user strategies from which user intentions are inferred.

6.1. Strategy Mining Method

The presented method was created by the authors based on the concepts and principles of the known mining techniques: the resource to be mined (news of the business world and the sales event log); the ETL (extract, transform and load) process, which allows the structuring of the KB and the QEL; the Python code, with which process mining is developed; the heuristic rules, which allows us to validate and weight the strategies extracted; the business expert, which allows us to infer the user intentions of the sales business information system.

In their works, Epure and Khodabandelou [5,7,8] developed intention mining methods based on the process mining and process-aware information systems techniques, and both authors centered their works on software developers.

6.2. User Strategies

For all businesses in general, the user strategies were stored in the KB (candidate_strategy data table), alternatively we gathered user strategies for a specific sales business (strategy_qel data table). The user strategies of the information system were validated and weighted through the heuristic rules. Huang, in his work [14], analyzed words and sentences online to infer the user intentions of web applications.

6.3. User Strategies Formalization

The possible business strategies and their characteristics were extracted from the KB for businesses in general. The strategies of a business user were extracted from a log of the activities of a specific business. The existence of the user strategies of the specific business was verified using the KB and its formalization according to the heuristic rules. In their work, Khodabandelou [3] promulgates map formalism model processes according to actor intentions and supports process variability by defining different strategies to achieve them. Therefore, confronted by a specific situation and a particular intention of an actor, the process model reveals the strategy to fulfill this intention. In our work, user intentions were inferred based on the strategies applied by the user to develop their business activities.

6.4. User Intentions

Without losing sight of the business’s rules and without affecting the current activities [17], users conduct their daily activities whilst trying to reach an optimal level of performance; often, information systems do not have enough flexibility to achieve these activities’ objectives. Therefore, users utilize their own strategies, with the intention of achieving their personal goals, which are always aligned with the business’s goals.

In the present work, and based on user strategies, we inferred their intentions (see Table 6) through the proposed mining method. Furthermore, these user intentions were weighted through compliance with the business’s heuristic rules. The user intentions can be interpreted by an expert within the business, which can help to infer user behavior.

6.5. Internal and External Threats to This Work

Based on the scope and limitations of this work, the following limitations are presented: the background and context are the business’s information systems, and their users and functions are the business’s policies. Consequently, this work is limited to business information systems.

The novel approach, “intention mining”, has shown little development; in our previous works [10,16], a literature review was developed, and the state of the art and the necessary data quality were defined.

The principal authors of the intention mining topic, such as Khodabandelou, Hug, Deneckère, Salinesi, Epure, and Brinkkemper, have oriented their works towards providing software developers with recommendations. In this work, intention mining was centered on the business information system user, which defines the scope of the present work.

The lack of current business data and information could be considered a limitation. However, real historical data were obtained to develop the present work.

6.6. Method Implications

The development of a new intention mining tool would have the following implications:

• it would be applied in the identification of the gap between a business’s current processes and actual processes;
• defining and structuring user intentions based on their strategies;
• defining levels of abstraction for the executed processes according to user intentions;
• ranking user intentions according to the levels of process abstraction;
• generating intentional process models according to the predefined process abstraction levels;
• intentional process model validation;
• Identifying user behavior, and based on it, providing recommendations for improving business information systems.

The method was effective and allowed us to obtain results for the sales business. The same can be generated for any other business, as long as the business has an event log.

7. Conclusions

In the absence of data on a specific business, the data used in the development of this research were acquired from free repositories (Signal Media, Google’s BigQuery platform, and cloud and Kaggle) and academic repositories (Stanford NPL Group). We structured a KB from flat files based on relational and dimensional database models. All of the resources used in this work are free to access and available in the GitHub repository [23]. However, the indispensable resources of the method are the KB and QEL (which provides a large number of resources to conduct mining), the Python code (mining tool), and the business’s rules (heuristic rules).

Due to the limited processing capacity available, of the million articles available in the NewIR’16 repository, a sample of 200 articles was used. It is evident that the obtained result is good, and we presume that this would improve with a greater number of articles.

To validate the activities of the specific sales business and the user strategies, a knowledge base was built for any business in general, so that, for the proposed mining method, there is a debugged and sufficient resource.

The results obtained can be used by business administrators to identify user behavior in the development of their activities, and based on this, to identify the requirements for improving the information system and the business’s current process.

This method was applied to a specific sales business, but it was generalized, so that the method can be applied to any business; accordingly, it is necessary to have the event log of the business and the judgment of an expert within the business.