Ontology of Social Service Needs: Perspective of a Cognitive Agent

Abstract

This paper proposes an ontology of social service needs for the evaluation of social service providers. Existing ontologies in the social service domain define metrics to evaluate the efficient use of resources by service providers. The ontology presented here represents service provisioning from the perspective of a cognitive and goal-driven client to evaluate services based on how well they remove a client's constraints and meet client needs. This ontology is grounded in real-life requests made by participants of a Housing First intervention program, resulting in 57 different goal types. Each goal is mapped to one or more basic human need defined by Maslow's Hierarchy, as inferred from the goal's type, the motivation behind it, and the client's demographics. Finally, as clients interact with service providers, three different types of goal orderings are required to capture goal ranking during the planning and execution phases. These include the client's preferred order, Maslow's hierarchy order, and the practical order imposed by the logistical constraints of service providers.

Keywords

• goals agent-based simulation cognition ontology human behaviour

1. Introduction

This paper proposes an ontology of needs for human-like agents that interact with a social service provisioning system. The ontology is based on data about the types of requests made by social service clients in a real-life intervention program. Existing ontologies focus on the process of service delivery, categorizing services and resources to ensure an efficient provisioning to incoming clients [ 1,15,11 ]. In the work proposed here, an ontology is created that allows for the evaluation of service provisioning from the client’s perspective. By identifying goals of clients and the services that satisfy them, it is possible to create a high-fidelity client emulation model for the purpose of social service evaluation [ 3 ]. Towards such a model, the ontology presented here provides competencies not yet provided elsewhere. The ontology is used to identify relationships between clients and service providers, including client needs, constraints, and motivations. The ontology also differentiates service-side concepts like resources, programs, and a metric for client outcomes. To support a cognitive agent, the ontology makes a destination between three different goal ranking used for goal reasoning and planning [ 3 ]. First, the ontology can be used to infer correct needs associated with Maslow’s hierarchy, by providing a set of domain-specific mappings between data provided by services and the hierarchy based on theoretical analysis of needs [ 10 ]. Second, a client’s own preferred ranking can be identified based on order requests are made on questionnaires and service request forms. Third, the practical ranking represents the order goals were actually satisfied in by services, as captured by service-side data, and takes into account environmental constraints imposed on the service provider.

Generally, human needs are difficult to capture. There are several theories that define motivation as “drives”, but these are too vague and inflexible to construct a computational model of a cognitive agent’s motivations and preferences [ 9 ]. Instead, goals are provided a priori and influence a goal-driven agent’s behaviour [ 3 ]. By evaluating the social service delivery process through data provided by participants in a real-life intervention program, an ontology is developed that captures the relation between client goals and the services they use.

There are several ontologies that capture social service provisioning from the provider’s perspective [ 1,15,11 ]. However, no ontology exists that focuses on client needs and motivations from the client’s perspective. At the same time, human motivations have long been credited with influencing decision making in the social service domain [ 2 ]. To assess a client’s current state, questionnaires such as the “Service Prioritization Decision Assistance Tool” (SPDAT) capture past and current needs. Once a client’s state and outstanding needs have been identified, techniques like Motivational Interviewing and Acceptance and Commitment Therapy are used to facilitate change in their behaviour that aligns with the clients motivating factors [ 2 ].

The proposed ontology provides an ontological representation for four aspects of social service client needs missing today. First, client needs made up of 763 requests found in the data are categorized into 57 different goal types. Each type is defined by the agent’s motivations, constraints, resources needed, and the services offering those resources. Second, the relation between a client and a service provider is based on the constraints faced by clients, not services. Third, each goal type has a homeless-specific mapping to one or more levels of Maslow’s hierarchy. Such mappings are not trivial, and the ontology infers appropriate mappings based on request types and client characteristics. Fourth, three goal orderings are identified for different phases of a client’s interaction with the provider. These include client preferences during the planning phase, Maslow’s order during plan execution phase, and practical ordering based on logistical constraints placed on the service provider.

2. Method

To capture how a service provider satisfies goals of clients, this paper develops the Ontology of Social Service Needs (OSSN). The ontology is developed using the ontology engineering method. Ontology engineering is a systemic way of constructing and evaluating an ontological representation of a domain [ 6 ]. First, motivating scenarios are identified to define the scope and objectives the ontology is meant to resolve. Second, a set of informal competency questions are defined which the ontology should answer. Third, an ontology is constructed that represents knowledge required to answer identified competency questions. Finally, the informal competency questions are translated into formal competency questions using the terminology and formal language that allows for the automation of querying identified questions. The work presented here represents the ontology in OWL Single Don’t know

Declined to answer Has your family situation changed since the last follow-up was completed? Yes

Don’t know

Declined to answer Which of the following best describes your current family situation?

Couple

Single parent family

Head of two-parent family

Other parent in two-parent family Are you pregnant?

Yes

No

Don’t know

Declined to answer How many dependents (under 18) do you have? (only include those also enrolled in the program) ________ Have you gained paid employment within the past 3 months? 2 The Calgary Homeless Foundation (CHF) has provided a dataset that captured informationDona’tbknoouwt clieDencltinsedatso atnhsweeyr participate in a “Housing First” (HF) intervention program adAre you currently attending employment related training? Yes - Full-time Yes – Part-time No Don’t know Declined to manisnweirstered by CHF. The CHF-HF dataset contains information on approximately 4,000 uHnavieqyuouecocmlpileetnedtsantehmapltoypmaenrttrieclaitepdatrtaeindingipnrotghraem wHithFin pthreopagstr3ammontihns?CalYgesary, NCo anaDdoan’tfkrnoowm 20D0ec9linetdoto2015. answer TArheeyoui ncufrroenrtmlyaattteinodinng fwurtahesr ecduoclaltieoncctlaesdses?usinYges -SFuPll-DtimAeT qYuese–sPtairot-tnimne aireNso. A cDoonm’tknpolwete dDeecslincerditpotainoswner of tIhNeCOdMaEta and analysis is provided in [ 5 ]. Based on the data, the ontology categorizes What are your current sources of monthly Retirement pensions, superannuation & clients according to fifteen key dCheildmTaoxgCrreadipt$h_i_c__s_.__S_PDAT also captures different client reincome (before tax)? (Check all that apply and annuities $________ qinudiecastetsamfoounrt) basic needs, as per FEimgpuloyrmeen1tI.nsPuraanrctei(cEIi)p$a__n__t_s__w_ere surveyed at program intake Self Employed $________ with follow-up interviews every tFhullr-teimee mEmpolonymthenst $u__n_t_i_l__e_xiting the program. By grouping 763

Aboriginal Funding $________ Student Funding $________ unique requests captured, 57 need categories represent goal types in the ontology.

Guaranteed Income Supplement or

Alberta Works/Income Support $________

2.2A.ssuMredoIntciovmaetfionrtghe SSecveerenlyarios

Handicapped (AISH) $________ Survivor’s Allowance $________

Housing Supplements $________ Long-term Disability (private) $________ Old Age Security Pension (OAS) $________ Other Tax Credits $________ $________

How to monitor client proPagrtr-teimsesE?mployment $________

Child Support/Alimony $________

How to monitor service delivery performance? MoBitninvinag/tRiencygclinsgc/BeonttlaerPiicokinsg f$o__r__t_h_e__ OSSN focus on the evaluation o$_f__s_o__c_i_al service policy from theCapnaedrasPpenesicontiPvlaen Boenfefcitsli$e_n__t_s__t_h_at use them. These include:

Canada Pension Plan Disability Benefits Other _______________ $________ How to evaluate interventPiaonhnandplinrgo$g__r_a_m___s_ in the social service space? No Income Don’t know Declined to answer The general approach to evaluating a program is to identify the percentage of clients who were successful [ 4 ]. The criteria for eligibility inHtoousiangpFirrsotFgorlloawm-up iAsssetshsmeenpt r(3o-6b0ambonitlhi)t-yPaagep2aofr3tic

Updated 7/27/2015 ipant will be successful based on their information at intake. With the HF program, it is not clear which cohorts will be successful [ 14 ]. Since simply relying on demographics is not sufficient, the motivating scenarios arise from the need to understand the interaction between clients and services as they participate in the program to meet their needs.

War Veterans Allowance/Veterans Benefits $________

Workers’ Compensation Benefit

2.3. Competency Questions

The focus of the competency questions for OSSN is to answer queries about the relationship between client needs and service providers captured by SPDAT questionnaires. Client questions address the three main concepts captured about clients, their needs, constraints, and demographics. For the complete list of questions see [ 5 ].

Q-1 Which demographic is asking for MH need X most? Q-2 Does client X ask for goals in the same order as client Y?

2The Calgary Homeless Foundation: http://calgaryhomeless.com/.

Q-3 What constraints clients with demographic X? Q-4 Are wrong conditional goals assigned to any client? Q-5 What services are needed together to address “childcare goals”? Q-6 What resources and service are needed to address a client’s security level needs? Q-7 How well do programs address physiological and security needs of clients? Q-8 Are resources available when needed? The first group is a sample of questions (Q-1 to Q-4) that examine the ontology’s ability to represent data provided in the CHF-HF SPDAT dataset. Focus is placed on the requests made by clients. This includes mapping the requests to Maslow’s hierarchy, capturing the order of requests, and associating them with possible motivations and constraints that prompted the requests. Using the provided demographics, OSSN infers the correct MH level to map participant requests to. The second group is a sample of questions (Q-5 to Q-8) that evaluate the ontology’s ability to capture services available to clients. By associating services with client constraints, the objective is to answer questions about service provisioning from the perspective of the client.

3. Engineering the Ontology of Social Service Needs

To engineer our ontology of needs, we first analyze how Maslow’s hierarchy can be applied to this domain to create a domain-specific mapping. We then identify high-level concepts required to categorize requests and present axioms included in the ontology.

3.1. Maslow’s Needs for Homeless Clients

While basic motivation for human needs is ill-defined [ 9 ], there is some consensus that behaviour models can rely on theories like Maslow’s hierarchy (MH) for grounding goals in basic human needs [ 10 ]. A need can be considered as a “master” goal, an innate requirement for an agent without a triggering activity. Such needs always exist with varying degrees of urgency. All other goals or sub-goals are regarded as tangible states that can be achieved and satisfied through a series of activities. MH categorizes tangible goals into five categories of basic human needs. While there is mostly consensus on the categories, there is less consensus on the correct order of MH levels and whether it can be applied universally across populations and cultures [ 13,7 ]. Generally, the first group of needs are short-term needs important to our survival. The second group includes longterm needs that serve to improve our life and society at large.

The mapping of goals to MH level needs is especially problematic for the homeless population. Mappings are conditional on a combination of demographics, goal types, and previously satisfied goals [ 12,7 ]. For example, housing (long term) and housing temp(orary) is only a physiological level need for absolutely homeless, and a security need for relatively homeless. Also, family needs are not necessarily a social level need. For example, when providing for a child’s needs, the goal is mapped based on the needs of the child. However, any motivations and constraints are those of the agent. Also, not all mappings are direct, one-to-one mappings between a need and an MH level, as discussed in section 4. Some span multiple levels at once, while others are spread across multiple levels to be satisfied in a sequence over an extended period of time. For example, requesting laundry services impacts a client’s self esteem, ability to socialize, and

SPDAT Request None, declined to answer Addiction support, case management, child care, education, employment training, family support, goods misc, life skills, referral, social Computer, counseling, debt reduction, disability support (for relatively homeless), education, employment training, family support, forms, goods family, goods infant, goods misc, hygiene, identification, laundry, life skills, money family, money planning, money social, phone (for non-elderly), referral, tenant insurance support, transportation Aboriginal, child care, computer, counseling, disability support (for relatively homeless), education, forms, health support, hygiene, immigrant services, laundry, life skills, miscellaneous support, money family, money social, phone (for non-elderly), referral, social, social family, transportation, utility arrears Advocacy help, advocacy legal, child care, clothing, counseling, disability support (for absolutely homeless), don’t know, forms, goods infant, goods misc, health support, housing (for relatively homeless), housing goods, housing maintenance, housing safety, housing supplement, housing temp (for relatively homeless), hygiene, identification, immigrant services, income, laundry, medication, mental issues, money goods, money health, moving, phone (for elderly), referral, rent arrears, rent shortfall/subsidy, security, security deposit, tenant insurance support, transportation, utility arrears Addiction support, food, furniture, home goods, housing (for absolutely homeless), housing temp (for absolutely homeless), mental issues prevents violence from others, hence is spans the esteem, social, and security levels. The final mappings for 57 goal types consolidated from the 18 request types captured by the SPDAT questionnaire section in Figure 1, including 745 entered by clients in the “other” fields, are provided in Table 1 with a complete analysis in [ 5 ].

3.2. Ontology of Social Service Needs

Based on the client requests captured in the CHF-HF dataset and directly, conditionally or unconditionally mapped to MH levels, the following ontological entities are represented. An agent’s relation to their goals and the services they use is represented by the Ontology of Social Service Needs (OSSN). This relation is comprised of its Maslow need and order ranking, followed by a concrete goal requested by a participant, personal motivation for that goal, and constraints preventing goals from being satisfied. The agent’s need is mapped to an MH level. Motivation is a description of why an agent might want to pursue this goal. It provides additional information for mapping a goal to the appropriate MH level. For example, “childcare” is a broad category of needs associated with the agent’s child’s needs. The motivation to keep a child out of harm’s way would associate a goal with the physiological level, as it prevents physical harm. This may include a request for emergency childcare and contacting child protective services. Child care may also be motivated by wanting to raise well-adjusted and social children and mapped to the agent’s esteem level.

The service provider is represented with resources and services that relieve an agent’s constraints. A constraint is a high-level summary of unsatisfied preconditions preventing an agent from achieving their goals. The preconditions are satisfied by social services that provide resources. For example, the constraint preventing an agent from providing toys or social activities for their children might be a lack of money or not State Demographic object property subClassOf createdBy Service offers Program prefAgent

AgentRankedGoal InterimGoal MotivationDescription describedMotiveFor

triggeredBy prefMH MHRankedGoal

RankedGoal

Goal hasGoal

Agent hasOutcome Outcome prefPractical

PracticalRankedGoal constrainedBy MHGoal mappedTo

MHNeed accessedBy forProgram

Constraint requiredBy

Resource knowing about available activities (i.e. lack of information). The service represents the service provider, program, or department that makes the resource available to the agent.

3.3. OSSN: Formal Definitions

This section provides the formal definitions for OSSN, represented in OWL syntax [ 8 ]. The OWL (Web Ontology Language) was chosen since it is one of the most common ontology languages on the Semantic Web. Main OSSN classes and properties are represented in Figure 2. Agents and Goals Clients are represented as human-like and goaldriven agents. Hence, the property hasGoal defines the Agent class as one that has at least one Goal state, as per Axiom 1. Axiom 2 defines the Goal class as a state triggered by some underlying MH need, but constrained from being true. The MotivationDescri ption class captures the agent’s expressed motivation for requesting a goal, as per Axiom 3.

Agent v 9hasGoal:Goal Goal v State u 9triggeredBy:MHGoal u 9constrainedBy:Constraint MotivationDescri ption v 9describedMotiveFor:Goal u 9ex pressedBy:Agent (1) (2) (3) Goal Constraints A goal state is constrained by the Constraint class, a state that summarizes unmet preconditions that prevent the goal state from being true. For example, an agent cannot buy food from a store if they do not have money. Having money is a precondition state that must be true before purchasing food. The state lackO f Money is a Constraint that prevents the Goal class moneyForFood from becoming true. For a state to be a constraint, it must also be resolvable by a resource. A non-resolvable constraint identifies an incorrect goal or action. For example, requesting legal advocacy from a housing worker describes an incorrect action if the goal is to find housing. Hence a

Constraint class is a State class that requires a Resource class (requiredBy :Resource)

and is actively constraining a Goal class (constrainedBy :Goal), as defined in Axiom 4.

Constraint v State u 9requiredBy :Resource u constrainedBy :Goal (4) MH Goals And Interim Goals Requests made by agents to satisfy expressed goals are triggered by an underlying MH level need associated with it. MHGoal represents such a need that triggers the requested goal. Each MHGoal is mapped to one or more MH levels. For example, while moneyForFood is a Goal, notBeHungry is the MHGoal state that triggers it. notBeHungry is then mapped to the “physiological” MH level. In OSSN, the triggeredBy property captures the relation between a requested Goal and its underlying MHGoal. The mappedTo property captures the relation between the MHGoal and its underlying MH level class MHNeed. These classes are defined in Axioms 5 and 6.

MHGoal v 9triggeredBy :Goal u 9mappedTo:MHNeed MHNeed 9mappedTo :MHGoal u fphysiological t security t

social t esteem t sel f Actualizationg InterimGoal v Goal u :8mappedTo:MHNeed (5) (6) (7) Finally, interim goals are sub-goals required to satisfy preconditions of actions that satisfy existing goals. For example, walking to the store to buy food is an interim goal. The InterimGoal class is defined as a subclass of Goal that is not mapped directly to an MH level, as defined by Axiom 7.

Agent Demographics An agent’s demographics are used to automatically infer conditional mapping. A conditional goal is a type of MHGoal class mapped to an MH level based on an agent’s Demographic class. For example, consider the examples in section 3.1. The goal of temporary shelter for agents in the “absolutely homeless” demographic is mapped to the physiological MH level. For agents in the “relatively homeless” it is mapped to the security MH level. A Demographic is a subclass of State class that defines the state of an agent, as per Axiom 8.

Demographic v State > v 8homelessState:fabs; relg AbsHomeless Demographic u homelessState : abs RelHomeless Demographic u homelessState : rel AbsHomelessAgent v Agent u AbsHomeless RelHomelessAgent Agent u RelHomeless ? v AbsHomeless u RelHomeless (8) (9) (10) (11) (12) (13) (14) Demographic properties define the actual “demographic” state true for the agent. For example, the following axioms define how to identify an agent as either absolutely or relatively homeless. First, the property homelessState in Axiom 9 has a range of “abs” and “rel” to represent an absolutely and relatively homeless status, respectively. Next, Axiom 10 defines the AbsHomelessState class as the intersection of the Demographic class and a class for which homelessState=abs. Similarly, Axiom 11 defines the RelHomelessState class as the intersection of the Demographic class and a class for which homelessState=rel.

Next, to assert that an agent is absolutely homeless, AbsHomelessAgent is the subclass of the intersection between the Agent and AbsHomeless classes, as defined in Axiom 12. For some agent A the assertion AbsHomelessAgent(A) categorizes A as an absolutely homeless agent. Its relatively homeless counterpart is defined in Axiom 13. Since absolutely and relatively homeless types are disjoint sets, having the same agent classified as both produced an inconsistent ontology, as per Axiom 14.

Service Provider and Resources The service provider is represented by the Service class. A service is something that can be accessed by an agent and creates resources, as defined in Axiom 15. For example, a “social worker” is a multi-functional service offered by a shelter. A social worker can provide a variety of resources, such as booking a bed, information about childcare, or finding a suitable mentor. It follows then, that the Resource class is defined as something a service creates and that is required by a Constraint class, as defined by Axiom 16.

Service v 9accessedBy:Agent u 9createdBy :Resource Resource v 9createdBy:Service u 9requiredBy:Constraint (15) (16) Program and Agent Outcome The last set of main classes OSSN supports are those that capture an agent’s outcome in a program that offers multiple services. An agent can access a service, but their outcome is evaluated in the context of the program. Hence, a Program class is defined as the intersection of classes that offer a Service and have an Outcome, as per Axiom 17. The Outcome class relates an agent’s status to a program, as per Axiom 18, with possible statuses as success, f ail, missing, or active.

Program

Outcome

9o f f ers:Service u 9 f orProgram :Outcome 9 f orProgram:Program u 9hasOutcome :Agent (17) (18)

3.4. Ranked Goals

Ranking goals allows a cognitive agent to reason about goals in terms of their importance to the agent [ 3 ]. A goal state can be preferred over another. If a preference is assigned to a goal it is considered a subclass of the RankedGoal class, with a unique ordering relation. A RankedGoal is any goal that has an integer preference assigned to it with the pre f data property, as defined by Axiom 19. However, goals can be ranked based on one of three order relations.

MHRankedGoal v RankedGoal u 9pre f MH : xsd : integer PracticalRankedGoal v RankedGoal u 9pre f Practical : xsd : integer First, during the planning phase, the agent uses their own preferred goal order to calculate the utility of each plan. The agent’s preferred ranking is represented by the AgentRankedGoal class as defined in Axiom 20. It is a subclass of the intersection between a RankedGoal, and a class with both pre f Agent and hasGoal relations. For example, given Goal states si and s j along with the assertions hasGoal(A; si), hasGoal(A; s j), pre f Agent(si; 1), and pre f Agent(s j; 2), the goal state si is preferred by agent A over s j.

During the plan execution phase, Maslow’s classical order is used to calculate the utility of goal state as actions to satisfy them are executed. The MH order is represented by the property pre f MH. A goal ranked by MH is an MHRankedGoal class as defined in Axiom 21. It is a subclass of the intersection between a RankedGoal and a class with pre f MH relation to an integer value. For example, the goal Food is an MHGoal mapped to the physiological MHNeed. The assertion pre f MH(Food; 1) would specify that the physiological level Food is mapped to is the most important. For each MH level, a specific ranking class that relates prefMH to the type of MHGoal it is triggered by: Finally, the practical ranking of goals represents the order in which goals were satisfied during plan execution. This order is observed in the outcome of a plan following its execution. The data property pre f Practical captures this relation, as defined in Axiom 22. The practical rank is captured by logging the execution of a plan. For example, the goals si and s j ranked by agent A above can be satisfied in reverse order. The assertions pre f Practical(si; 2) and pre f Practical(s j; 1) capture this order.

4. Mapping CHF-HF Data to OSSN

An application of OSSN is to infer the mapping of requests captured by CHF-HF data in using an ontological representation. All recorded requests were combined into 57 basic needs associated with one or more levels of Maslow’s hierarchy. A sixth level was added for non-answers like “Don’t know”. The entire mapping between CHF-HF basic needs and MH levels is provided in [ 5 ]. The following sections provide ontological definitions required to map goals directly, conditionally, or to multiple MH levels.

4.1. Mapping Direct Goals In OSSN

Direct-mapping goals are those directly associated with a single MH level. Consider the following OWL examples of clothing and advocacy needs. A request made for an article of clothing is directly mapped to the security level, as defined by Maslow [ 10 ], hence a request for clothing is the expressed goal and MH goal mapped to the security MH need. The agent’s motivation for clothing is simply to “be clothed.” The concrete goal requested is to get “help with buying or receiving clothing.” The constraint faced by an agent is “lack of money.” The resource where an agent can receive information about obtaining clothing without money is a “charity.” Finally, the service offered by the charity that provides clothing is a “donation centre.” As a direct mapping, any goals of type GoalClothing are mapped to the same security level. Hence, any MHGoal triggered by a GoalClothing type is equivalent to a security class, with no other properties required, as per Axiom 24.

MHGoalClothing

MHGoalSecurity u 9triggeredBy :GoalClothing (24)

4.2. Mapping Conditional Goals In OSSN

Conditional goal-mapping requires some agent specific condition to identify which MH level a requested need is mapped to. Unlike the directly mapped goals for clothing, conditional mappings are inferred from the intersection of an agent’s demographic and their specific need. Consider a request for “temporary housing” at some shelter. Such requests are categorized differently for absolutely and relatively homeless clients. For absolutely homeless it is a physiological MH need, while for the relatively homeless it is a security MH need. In OSSN an agent’s homeless state is a demographic defined by Axioms 12 and 13 for absolutely and relatively homeless respectively. For both types of homeless agents, the MH goal is to find “temp housing shelter” motivated by wanting “temporary housing for a short time.” The requested goal is “get help to find temp housing.” The constraint faced by the agent is not knowing which beds are available and in which shelters. The resource is a temporary bed available at a shelter. The service is a social worker that provides information about the bed. Mapping the MH goal to an MH level is inferred from the agent’s homeless state and goal type, as per Axioms 25 to 29.

GoalForAbsHomeless v 9hasGoal :AbsHomelessAgent (25) MHGoalTempHousingPhysiological v MHGoalPhysiological u (26) 9triggeredBy :GoalForAbsHomeless u 9triggeredBy :GoalTempHousing MHGoalTempHousingPhysiological v MHGoalPhysiological u (27)

MHGoalPhysiological

MHGoalTempHousingSecurity v MHGoalSecurity u (28) 9triggeredBy :GoalForRelHomeless u 9triggeredBy :GoalTempHousing MHGoalTempHousingSecurity v MHGoalSecurity (29) First, an absolutely homeless goal class GoalForAbsHomeless is any goal that is requested by an absolutely homeless agent, as per Axiom 25. Second, a request for temporary housing, say getTempHousing2, is asserted as GoalTempHousing(getTempHousing). Mapping this goal to the physiological MH level is conditional on the agent being absolutely homeless as per Axiom 26. The MHGoalTempHousingPhysiological class, as per Axiom 27, is also defined as the subclass of MHGoalPhysiological. For relatively homeless agents, temp housing goals are mapped to the security level, as per Axiom 28. Similarly to the physiological goal in Axiom 27, the MHGoalTempHousingSecurity class is also defined as the subclass of MHGoalSecurity in Axiom 29.

4.3. Mapping Unconditional Goals In OSSN

Many OSSN needs are mapped to multiple MH levels at once. For example, doing laundry is mapped to security, social, and esteem MH level needs. Laundry is a request that impacts at multiple MH level needs, mainly security, social, and esteem. Each is mapped to the same MH goal to “feel safe with others,” as per the assertions in Axioms 30 a to c. The constraint faced by the agent is that they do not have money to pay for their own laundry. The resource is the free laundry facility they can access. Finally, the service provider is a shelter that is offering free laundry service.

MHGoalLaundrySecurity( f eelSa f eWithOthers)

MHGoalLaundrySocial( f eelSa f eWithOthers)

MHGoalLaundryEsteem( f eelSa f eWithOthers)

(30a) (30b) (30c)

5. Discussion

The OSSN provides an ontological representation of a client’s motivations, goals, and different ways goals are ranked. The focus is placed on how the service can relieve constraints exhibited by the agent, which resources are required, and which services provide those resources. The service provisioning is not centred around service efficiency, but on satisfying the underlying constraints faced by clients. To this end, the CHF-HF dataset captures client needs as they participate in the housing first intervention program. Since needs were collected every three months, the data also captures how a client’s needs change over time. By identifying three different goal orderings, changing order of goals and their rankings can be represented and used for goal reasoning by a cognitive agent. Depending on the agent’s demographics, OSSN infers how goals should be mapped to Maslow’s hierarchy.

Following the ontology engineering method, motivating scenarios proposed in section 2.2 identify the scope and focus for the development of OSSN. Competency questions identify issues that should be addressed and what vocabulary is required to answer them. For lack of space, the complete results and analysis are presented in [ 5 ]. Overall, the ontology performs well on questions that relate to client and service types. The relationship between clients and goals is well represented, where SPARQL queries are able to ask and answer questions about demographics and goals. OSSN is also capable of answering queries about service provisioning. By relying on the Outcome class, OSSN can answer some queries that relate to the progress participants make in a program. OSSN has several limitations. Any questions with a temporal dimension are not supported by OSSN. For example, the rate at which resources are used or when they become unavailable cannot be answered by OSSN.

6. Related Work

Several ontologies overlap with the proposed ontology and address some of the competency questions. These, however, are service-oriented, focusing on modelling processes and constraints of the service provider rather than the impact on client outcomes. The Open Eligibility Project (OEP) is a taxonomy of service categories offered to clients [ 1 ]. The agent is represented by the “human situations” category. It includes age group, citizenship status, criminal history, disabilities, health, household, and urgency. However, each term lacks a definition leaving them open to interpretation. For example, emergencies are simply qualified as “In Crisis,” “In Danger,” or “Emergency.” The GCI ontology focuses on housing and classifies clients as absolutely or relatively homeless [ 15 ]. The resources available to the clients are different types of housing. The competency questions GCI addresses focus on details about specific households and aggregate information about city resources and household types. For example, GCI can answer who the individuals in a particular household are and whether that household is considered a “slum household.” The INSPIRE ontology captures processes and resources of service providers focusing on elderly and adults living with disabilities [ 11 ]. Client needs can be categorized as physical or social, or a combination of the two, along with an urgency indicator. This is used to efficiently identify the appropriate department to transfer a client. The competency questions INSPIRE can answer focus on service assignment. Services and internal workflows are well represented, while client needs and underlying symptoms are not.

7. Conclusion and Future Work

Up to now, the client’s perspective of social service policy evaluation has been missing. The work presented here fills this gap by providing an ontological representation of a client’s motivations, goals, and different ways goals are ranked. The Ontology of Social Service Needs (OSSN) identifies the semantic relations between requests made by a client to a service provider, based on data provided by real-life clients about their changing needs while participating in a real intervention program. The ontology provides a goal ranking used by cognitive agents to prioritize goals while planning their actions. The ontology was evaluated by answering certain competency questions. The questions that were not answered are the basis for future work. This involves goal reasoning and planning to simulate a client’s interaction with service providers.

Acknowledgements

This research was funded by NSERC Discovery Grant. The authors would like to thank the reviewers for their insightful comments, as well as the Calgary Homeless Foundation for the dataset, and state that the our findings do not reflect the views of the Foundation.

References

[1] Aunt Bertha Inc . http://openeligibility.org/, Accessed: May 25, 2018 .

[2] J B Bricker and S J Tollison. Comparison of Motivational Interviewing with Acceptance and Commitment Therapy: A conceptual and clinical review . Behavioural and cognitive psychotherapy , 39 ( 5 ): 541 - 559 , 2011 .

[3] Bart Gajderowicz , Mark S Fox, and Michael Gru¨ninger. General Model of Human Motivation and Goal Ranking . In 2017 AAAI Fall Symposium Series on Standard Model of the Mind, page 6 , Arlington , VA , 2017 . AAAI Press.

[4] Bart Gajderowicz , Mark S Fox, and Michael Gru¨ninger. Requirements for Emulating Homeless Client Behaviour . In Proceedings of the AAAI Workshop on Artificial Intelligence for Operations Research and Social Good, page 7 , San Francisco, CA, 2017 . AAAI Press.

[5] Bart Gajderowicz , Mark S Fox, and Michael Gru¨ninger. Report on the Ontology of Social Service Needs: Working Paper . 2018 .

[6] Michael Gru ¨ninger and Mark S Fox. Methodology for the Design and Evaluation of Ontologies . In IJCAI Workshop on Basic Ontological Issues in Knowledge Sharing , 1995 .

[7] Benjamin F Henwood , Katie-Sue Derejko , Julie Couture, and Deborah K Padgett. Maslow and Mental Health Recovery: A Comparative Study of Homeless Programs for Adults with Serious Mental Illness . Administration and Policy in Mental Health and Mental Health Services Research , 42 ( 2 ): 220 - 228 , 2015 .

[8] Pascal Hitzler , Markus Kro¨tzsch, Bijan Parsia, Peter F Patel-Schneider , and Sebastian Rudolph. OWL 2 Web Ontology Language Primer (2nd Edition) , 2012 .

[9] Paul R. Kleinginna and Anne M. Kleinginna . A categorized list of motivation definitions, with a suggestion for a consensual definition . Motivation & Emotion , 5 ( 3 ): 263 - 291 , 1981 .

[10] Abraham Harold Maslow . A theory of human motivation . Psychological Review , 50 ( 4 ): 370 - 396 , 1943 .

[11] Elaheh Pourabbas , Antonio D'Uffizi, and Fabrizio L Ricci . A Conceptual Approach for Modelling Social Care Services: The INSPIRE Project . Data Integration in the Life Sciences , pages 53 - 66 , 2017 .

[12] John R Sumerlin. Adaptation to Homelessness: Self-actualization, Loneliness, and Depression in Street Homeless Men . Psychological reports, 77 ( 1 ): 295 - 314 , 1995 .

[13] Andrew B. Trigg . Deriving the Engel curve: Pierre Bourdieu and the social critique of Maslow's hierarchy of needs . Review of Social Economy , 62 ( 3 ): 393 - 406 , 2004 .

[14] Jennifer S Volk , Tim Aubry, Paula Goering, Carol E Adair, Jino Distasio, Jonathan Jette, Danielle Nolin, Vicky Stergiopoulos, David L Streiner, and Sam J Tsemberis. Tenants with additional needs: when housing first does not solve homelessness . Journal of Mental Health , 8237 (December): 1 - 7 , 2015 .

[15] Yetain Wang and Mark S Fox . A Shelter Ontology for Global City Indicators (ISO 37120) . 2015 .