an intelligent internet shop-assistant recognizing...

14
Recent Advances in Intelligent Information Systems ISBN 978-83-60434-59-8, pages 13–26 An Intelligent Internet Shop-Assistant Recognizing a Customer Personality for Improving Man-Machine Interactions Adrian Horzyk, Stanislaw Magierski, and Grzegorz Miklaszewski Department of Automatics, AGH University of Science and Technology, Cracow, Poland Abstract Turnover in Internet commerce is growing rapidly. Internet shops are increasingly better in recognizing needs of their customers. The main area of interest is cus- tomer preferences, which predictably can be based on purchase history. There is also a second area of the customers needs, which is not managed so far, and that is a customer personality and needs entailed it. The goal of this paper is to present successful implementation of the recognition of customer personalities, which has an impact on improving human–computer interactions. For that purpose, a sample Internet shop was created with a chatbot playing the role of a shop-assistant that is similar to a real shop-assistant in a traditional shop. The chatbot performs a trade conversation with a customer through an in-built natural language process- ing unit. In a conversation, it tries to figure out desirable product preferences of customers and it analyzes words, phrases and sentence constructions of customers to automatically indicate their personality in order to correctly react to their pre- dicted personality needs. This paper, based on a recognized personality, describes how the chatbot can adjust its model of actions in the following fields: the display of product presentation and the choice of phrases used in conversations. Keywords: automatic human personality recognition, adjusting of chatbot reac- tions to human needs, chatbot, psychology, artificial intelligence, computer lin- guistic 1 The Important Business Need The interest in commercial use of Internet grows continuously. According to many reports [Grzechowiak (2007)], each year brings on average about 20%-35% increase in on-line sales. Online shopping grows much faster than traditional ones and there are several reasons for this. The Internet cuts off a long chain of middlemen, giving customers direct access to vendors; transaction costs are decreased; and Internet shopping is more convenient (as customers have an access to Internet stores 24/7 from anywhere in the world). On the one hand, the Internet provides a cheaper and more convenient way of trading; however it poses various concerns when it comes to interface and security issues. Generally, online shops are self-service applications with a web interface.

Upload: duongtram

Post on 15-Aug-2018

224 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Recent Advances in Intelligent Information Systems

ISBN 978-83-60434-59-8, pages 13–26

An Intelligent Internet Shop-Assistant

Recognizing a Customer Personality for

Improving Man-Machine Interactions

Adrian Horzyk, Stanis law Magierski, and Grzegorz Miklaszewski

Department of Automatics, AGH University of Science and Technology, Cracow,Poland

Abstract

Turnover in Internet commerce is growing rapidly. Internet shops are increasinglybetter in recognizing needs of their customers. The main area of interest is cus-tomer preferences, which predictably can be based on purchase history. There isalso a second area of the customers needs, which is not managed so far, and that isa customer personality and needs entailed it. The goal of this paper is to presentsuccessful implementation of the recognition of customer personalities, which hasan impact on improving human–computer interactions. For that purpose, a sampleInternet shop was created with a chatbot playing the role of a shop-assistant thatis similar to a real shop-assistant in a traditional shop. The chatbot performs atrade conversation with a customer through an in-built natural language process-ing unit. In a conversation, it tries to figure out desirable product preferences ofcustomers and it analyzes words, phrases and sentence constructions of customersto automatically indicate their personality in order to correctly react to their pre-dicted personality needs. This paper, based on a recognized personality, describeshow the chatbot can adjust its model of actions in the following fields: the displayof product presentation and the choice of phrases used in conversations.

Keywords: automatic human personality recognition, adjusting of chatbot reac-tions to human needs, chatbot, psychology, artificial intelligence, computer lin-guistic

1 The Important Business Need

The interest in commercial use of Internet grows continuously. According to manyreports [Grzechowiak (2007)], each year brings on average about 20%-35% increasein on-line sales. Online shopping grows much faster than traditional ones and thereare several reasons for this. The Internet cuts off a long chain of middlemen, givingcustomers direct access to vendors; transaction costs are decreased; and Internetshopping is more convenient (as customers have an access to Internet stores 24/7from anywhere in the world).

On the one hand, the Internet provides a cheaper and more convenient way oftrading; however it poses various concerns when it comes to interface and securityissues. Generally, online shops are self-service applications with a web interface.

Page 2: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

14 Adrian Horzyk et al.

The interface, though are generally easy to use, does not allow the kind of personalinteraction that one gets actually when he physically visits a shop. There is also aproblem with trust [Egger (2003)]; concern has been voiced by many users over thede-humanized transaction system, with reservations around the process of givingout vital information such as credit card details.

As the user interface is the only way for an Internet business to interact andinfluence a customer, it is bound and required to have multiple marketing and salesattributes. Some of the more popular areas emphasized are product placement,advertising, customer support, video or audio support to assist understanding ofthe system, building trust through trust seals, and sale offers etc. A well designed,easy to use and customized interface to user needs is in demand. Human-ComputerInteraction (HCI) research field is increasingly gaining momentum and highly usedin the Internet industry.

Through testing, interface design has proven to be a great factor for salesconversion and helps to secure more sales. Some examples of design elements thatinfluence sales are composition, color balance, navigation placement and showingthe function of an application, which is relevant to an actual state. Another,but less popular, interface element is the personal assistant, which has a positiveimpact on the customer’s purchasing decision [Senger (2002)].

The general rule about web interferences was introduced by Jakob Nielsen,”Studies of user behavior on the Web find a low tolerance for difficult designs orslow sites. People don’t want to wait. And they don’t want to learn how to usea home page. There’s no such thing as a training class or a manual for a Website. People have to be able to grasp the functioning of the site immediately afterscanning the home page-for a few seconds at most.” [Nielsen (2000)]

2 Improvement of an Online Shop Interface

Innovation and testing in the Internet based company is a regular and ongoingprocess. The most advanced and modern e-shop needs to have a built-in abilityto adapt to customers’ needs. Every action by the customer is monitored andanalyzed to understand needs and elements that influence sales, so it can thusbe incorporated into the design. Based on this information an e-shop can bedynamically adjusted to help a customer to discover another products, so the web-interface should be optimized upon individual requirements. The primary exampleof an e-shop with this functionality is Amazon.com. Other modern e-shops enrichtheir customer experience by providing shop-assistance chatbots, which facilitatean access to shop inventory though natural language communication. An exampleof a popular online shop with a chatbot is IKEA.com with its simple Anna chatbot.However, both of these examples do not cater to rich customer response.

This paper describe a next step in evolution of web-based interfaces of onlineshops - the ability to respond and adopt interaction between a customer andan online shop based on a customer personality (Fig. 1). This paper is basedon researches and software created for MSc thesis “Self-adaptive e-shop operatedby a personal assistant chatbot based on CRM postulates” at AGH University ofScience and Technology 2009 and on a human intelligence and personality research

Page 3: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 15

Figure 1: A customer and an online Internet shop interface.

conducted by Adrian Horzyk - the co-author of this paper. The created softwareprovides a linguistic intelligent chatbot developed in Ruby on Rails, which is ontop of the open source e-commerce platform [Substract (2009)]. Thus, the chatbotis an embedded part of an existed e-commerce platform. The chatbot is domainspecific, it is developed to carry out a trade talk about assortment of an e-shopand to handle a discussion around the customer support domain. The createdchabot works as a state machine: each customer’s statement invokes a change ofa state and produces an outcome in a form of text response.

Each customer statement is processed by a morphology transformation, whichtransforms each token of a customer statement into a language base form. Basedon the base form and on the actual context, an appropriate statement pattern isassigned (Fig. 2). Based on this assigned pattern and the actual state of a chatbot,an appriopriate chatbot response is produced. At the end of this circuit, chatbot’sresponce is read by speech synthesizer IVONA [IVONA (2009)]. CLP libraryis used for the morphology transformation. CLP is a library which generates aninflectional Polish language dictionary. CLP was developed in Computer LinguisticGroup, led by Prof. dr hab. Wies law Lubaszewski at AGH University of Scienceand Technology, Cracow, Poland [Lubaszewski (1993)].

3 Personality Recognition

Human Personality Types (HPTs) widely introduced in the other paper Horzyk(2009) describe specific human actions and reactions dependent on a human per-sonality. The HPTs also figure out various needs connected with a personality.Moreover, the HPTs can give relevant answers to questions how to individuallymanage (Tab. 5) a contact with a customer to give him more satisfaction whentalking to an intelligent chatbot system.

This psycholinguistic model of human personality consists of 11 human person-ality types (HPTs) that represent special groups of human behaviours and humanpersonality needs (HPNs): Dominating - DOM, Maximalist - MAX, Inspiring -INS, Discovering - DIS, Verifying - VER, Systematic - SYS, Assurant - ASS, Har-monious - HAR, Emphatic - EMP, Task-Oriented - TAO and Balancing - BAL.

Page 4: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

16 Adrian Horzyk et al.

Table 1: The needs and likes for the HPTs.

Page 5: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 17

Figure 2: The psycholinguistic chatbot engine.

Table 2: The dislikes for the HPTs.

Page 6: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

18 Adrian Horzyk et al.

Table 3: The words and phrases which enable to recognize the HPTs.

Page 7: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 19

Table 4: The inflection, sentence constructions and topics which enable to recognizethe HPTs.

The ability to recognize HPTs and their needs allows a chatbot engine (Fig. 2) toformulate and use appropriate algorithms to conduct negotiations, talks or moreprobably persuade or ask somebody to do something. Intelligence can supportand reinforce or suppress and temper HPT actions and reactions and lets a humanto much more efficiently and easily fulfill needs, avoid increase of their intensitiesor decrease of their fulfillment. Intelligence can also make a human to behaveindependently of personality and physiology. Each HPT triggers a will to behaveas specified in table 1. Each HPT does not like, refuses or escapes things or ac-tions described in table 2. The HPTs can be recognized using words, phrases, aninflection and sentence construction described in tables 3-4 and should be treatedin the way described in table 5 to trigger positive personality reactions.

4 Conversation Adjustment

Each HPT has characteristic words and phrases (Tables 3-4), which are more fre-quently used than in the other types. The list of that words and phrase wereelaborated based on [Horzyk (2009)]. A conversation with a chatbot using a nat-ural language is very convenient place to notice HPT symptoms embedded in thespecific way of talking and in the usage of the specific words and phrases (Tab.3). A chatbot in an e-commerce system, like an online shop, is an irreplaceableopportunity to get to know customer thoughts and their ways of thinking. Thismay be a possible gateway to new opportunities for e-commerce innovations. Tomeasure variables of customer personality for each HPT, the determinant wordsand phrases are defined (Tab. 3). Additionally, each word or phrase is weighted.

In the custom created shop, recognizing personality is based on the occurrenceof characteristic words and phrases in customer statements. Each occurrence ofwords and phrases assigned to one or more HPT is multiplied by an assigned weightand saved into the personality table (PT). The decisive HPTs have maximal values,

Page 8: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

20 Adrian Horzyk et al.

Table 5: The profitable and beneficial treatment of the HPTs that value them.

which are greater than an average value of all PT values. The maximal values ofthe HPTs in the case shown in table 7 point a Systematic-Dominant customer out.

When the personality of a customer is determined, an e-shop chatbot changesits behavior to please customer specific psychological personality needs like:

• the will to make own decision or choice,

• the will to harmonize with people and things,

• the will to systematize and order products and their features,

• the will to find inspiration in something,

• the will to reveal and discover something,

• the will to give him a possibly long guarantee.

In the custom created shop, there are two ways of adaptation for recognition ofcustomer personality: the chatbot response phrases and the product presentation.

Page 9: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 21

Table 6: The example of operationalization determinants for Systematic Character.

The words and phrases for the HPT - Systematic Weigthfirst, second, third, fourth, fifth, next, 1

last, at last, now, earlier, later,at the beginning, start, start with,at the end, mess, tidiness, untidy,

gradually, step, one stage, arrange,classify, cluster, enumerate, list,

layout, compose, composition, group, structure,model, think, lay out, plan out, unfold, divide,

after the other spread out, time, on time, 2date, deadline, count, map schedule, chronology

appointment, timetable in order, order, sort, sequence,rank, level,

systematize organization, organize, 3in steps, one by one

Table 7: The example of the personality table filled out after a chatbot talk.

HPT Assigned pointsHarmonious 4

Assurant 1Systematic 21

Differentiating 3Discovering 9

Inspiring 1Maximalist 4

Dominant 21

This paper focus on describing the HPP adjustment for the product presenta-tion part of conversation. The product presentation happens on the end of theconversation, when customer requirements are collected.

Each of the featured HPT has assigned a certain set of these variables withvalues. Each set fully determines how products are presented to a certain customer.

5 Test scenario

This paper describes an innovative use of HPTs recognized in human-computer in-teraction (HCI) in an e-commerce environment. There is a strong need to measurethe impact, which the application using HPT recognition and adjustment has ona HCI interaction process and a final decision and satisfaction of customers. Theone way of evaluating HCI is by measuring user interface. There are many userinterface metrics available Bevan (2001). One of universally recognized standard

Page 10: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

22 Adrian Horzyk et al.

Table 8: Presentation of product is described through seven separable variables.

Adaptation variable DescriptionRegularity determine whether products parameters

should be presented in systematic order (a table)or in more inspirational form.

Sensuality determine an amount of relatedpictures and sounds to appear with a product.

Choice determine a number of proposed productsin a final offer

Extra determine an amount of extra informationabout a product (like history, famous owners, appearancein Hollywood productions and other curiosities)

Technical determine how many technical detailsare mentioned in an offer

Assure determine whether an information about guaranteeshould be presented with a product description

Figure 3: The example of the adjusted presentation of the products offer (left: for theInspiring personality type, right: for the Systematic personality type).

is ISO 9241 part 11: “Usability Guidance”, where the “usability” term is definedISO (1998):

Usability: Extent to which a product can be used by specified users to achievespecified goals with effectiveness, efficiency and satisfaction in a specified con-text of use.

where:

Effectiveness: Accuracy and completeness with which users achieve specified

Page 11: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 23

Figure 4: The controlled experiment scenario.

goals.

Efficiency: Resources expended in relation to the accuracy and completenesswith which users achieve goals.

Satisfaction: Freedom from discomfort, and positive attitudes towards the useof the product.

5.1 A Way of Evaluation

The metric used for the user interface evaluation was User Satisfaction (whichdirectly relate to ISO 9241 definition of Satisfaction ISO (1998)):

Satisfaction: Satisfaction measures the extent to which users are free from dis-comfort, and their attitudes towards the use of the product. Satisfaction canbe specified and measured by subjective rating on scales such as discomfortexperienced, liking for the product, satisfaction with product use, or accept-ability of the workload when carrying out different tasks, or the extent towhich particular usability objectives (such as efficiency or learnability) havebeen met...

The aim of this paper is to prove that the application of HPT adjustment tothe software interaction with a user process make sense. In the presented case,apart of adjusted chatbot responses, the biggest outcome of HPTs in use its finaloffer presentation. The final offer presentation is the stage in a conversation,which belongs on the HPT recognition the most. This is the last stage, where auser has only one action to perform, and that is to buy or leave. That is whyother ISO9241’s subscales like Effectiveness and Efficiency were not taken intoconsideration since that stage of a conversation does not affect them.

There are many ways of measure user satisfaction subscale. The chosen one issubjective measures of satisfaction, which is produced by quantifying the strengthof a user’s subjective opinions. Surveys is a kind of a research instrument whichare widely used in usability testing Kirakowski (1993) and are recommended byISO 9241 standard. The survey to conducted research was created based on Liker

Page 12: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

24 Adrian Horzyk et al.

psychometric scale. Survey was consist of question on user satisfaction, and foreach question there were 5 response categories: Strongly Agree, Agree, Don’t Know,Disagree and Strongly Disagree. Question were based on industry standards waysto evaluate a satisfaction factor Kirakowski (1993). Some examples of questionsare depicted below:

• I would recommend this software to my colleagues.

• I enjoy my sessions with this software.

• Working with this software is satisfying.

• The way that system information is presented is clear and understandable.

• This software seems to disrupt the way I normally like to arrange my work.

• Working with this software is mentally stimulating.

• There is never enough information on the screen when it’s needed.

• I think this software is inconsistent.

• I am willing to come back to the e-shop later.

• Using this software is frustrating.

• There have been times in using this software when I have felt quite tense.

• I think this software has made me have a headache on occasion.

• The software has a very attractive presentation.

• This software is really very awkward.

• I feel comfortable using this software.

• The organization of information on the screens is clear.

• The interface of this e-shop is pleasant.

• Overall, I am satisfied with this e-shop.

• I purchased my desire product.

• the shop assistant was very helpfull.

• Shop assistant smoothly guide me through shops’ inventory.

• I am happy with my order.

• I would not like to use this software every day.

5.2 Controlled Experiment

The experiment (Fig. 4) was arranged with two samples to evaluate the HPTfactor and emphasize its impact on a controlled experiment Wikipedia (2009):

Sample 1 (a control group): E-shop with a chatbot without HPT recognitionand adjustment.

Sample 2 (an experimental group): E-shop with a chatbot with HPT recog-nition and adjustment.

Each group consisted of 12 people and everyone had the same task to perform inan Internet shop: “Buy a mobile phone of your choice”. People who took partin the experiments were mostly students, friends and colleagues of this paper’sauthors. The experiment was conducted separately in a home environment on

Page 13: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

Shop-Assistant Recognizing a Customer Personality 25

the same portable computer. During the experiment, respondents were assistedby a researcher, whose only role was to give a printed survey after a person hasperformed the task. The average experiment session time took between 10 and 15minutes.

5.3 Results and their Processing

The lowest answer (“Strongly Disagree”) was evaluated as 1, the highest one(“Strongly Agree”) as 5. The total amount of points gathered in the controlgroup was used as a normalization factor for the experimental group. The totalamount of points of the control group was represented as 100% on the percentagescale. Then summarized points of the experiment group were converted to thepercentage scale with the final result of 126%. In other words, this experimentproved that the implemented HPT recognition method has a successful impact onuser satisfaction factor (1) and (2).

SatisfactionFactor =ExperimentalGroupPoints

ControlGroupPoints(1)

SatisfactionImprovement = 100% ∗

{

ExperimentalGroupPoints

ControlGroupPoints− 1

}

(2)

6 Conclusion

This paper described the Internet shop application that uses a novel approachto human-computer interfaces (HCI) based on recognition capabilities of humanpersonality types (HPTs) widely introduced in Horzyk (2009). The adaptation ofHPTs in the Internet shop for selling mobile phones has shown that the abilityto recognize the personality of a customer is crucial for development of Internete-commerce services and for achieving more satisfaction of their customers.

The role of an interface is significant in today’s Internet business. The con-trolled experiment presented in this paper proves that implementation of the self-adaptive mechanism considerably improved customer satisfaction. The test resulthas shown that the presented implementation of personality recognition has a sig-nificant impact for Internet business. There was observed that there was almost26% rise of an average satisfaction rate in the experimental group. The environ-ment where testing were run did not allowed the testing of a conversation (actualnumber of purchases) rate in Internet business. But it can be assumed that theconversation rate will raise together with customer satisfaction, which means anextraordinary opportunity for Internet business. This paper described a small steptoward improving intelligence of human-computer interfaces by adaptation of theHPTs and it opens the door to further intensive research in this field.

Page 14: An Intelligent Internet Shop-Assistant Recognizing …iis.ipipan.waw.pl/2009/proceedings/iis09-04.pdf · An Intelligent Internet Shop-Assistant Recognizing a Customer Personality

26 Adrian Horzyk et al.

References

Nigel Bevani, International Standards for HCI and Usability, International Journal ofHuman Computer Studies, 55(4), 533–552, http://www.nigelbevan.com/papers/HCI-Usability standards.pdf.

F.N. Egger (2003), From Interactions to Transactions: Designing the Trust Experiencefor Business-to-Consumer Electronic Commerce, PhD Thesis, Eindhoven University ofTechnology, The Netherlands.

M. Grzechowiak, P. Jarosz, Raport Ecommerce 2007, Internet Standard, Micha l Grze-chowiak, Internet Standard, http://www.Internetstandard.pl/news/143692 2/ Inter-net.Standard.i.Sklepy24.pl.prezentuja.raport.E.commerce.2007.html

A. Horzyk, R. Tadeusiewicz, A Psycholinguistic Model of Man-Machine InteractionsBased on Needs of Human Personality, LNCS, (in the publishing process).

J. Kirakowski and M. Corbett (1993), SUMI: the Software Usability Measurement In-ventory, a. J. Ed. Technol. 24.3, 210–214.

Enrico Senger, Sandra Gronover and Gerold Riempp, Customer Web Interaction: Fun-damentals and decision tree, Institute of Information Management, University of St.Gallen

Wies law Lubaszewski, Marek Gajecki, Piotr Pisarek, Pawe l Pietras, Micha l Korzy-cki, Krzysztof Dorosz, Henryk Wrobel, Alicja Orzechowska, Teresa Rokicka, DorotaKorwin-Piotrowska, Izabela Gatkowska, Barbara Moskal, Wiktor Dernowicz, Miros lawGajer, Adrian Horzyk, http://winnie.ics.agh.edu.pl/

ISO 9241 part 11

http://www.informationweek.com/773/web.htm

http://en.wikipedia.org/wiki/Experiment

http://code.google.com/p/substruct/

http://www.ivona.com