Multimodal spatio-temporal valuation system for optimizing real property use and development potential

Abstract

A system and method for a multimodal, spatio-temporal automated valuation model (AVM) to assess the market value and optimal use of real property is disclosed. The system includes multimodal input unit, a machine learning engine and an optimizer. The disclosure leverages heterogeneous multimodal inputs, including structured property metadata, narrative property descriptions, digital images (interior, exterior, aerial), historical sales records, zoning bylaws, official and secondary plan data, and geographic information system (GIS)-integrated planning layers capturing geospatial proximity to infrastructure, amenities, and environmental constraints. The system employs large language models (LLMs) to extract structured features from unstructured text and image data, and integrates these features with geospatial planning data. The machine learning engine includes deep learning temporal architectures to capture historical patterns across multimodal inputs, enabling prediction of both current market valuations and future appreciation scenarios under various planning and policy changes.

Description

MULTIMODAL SPATIO-TEMPORAL VALUATION SYSTEM FOR OPTIMIZING REAL PROPERTY USE AND DEVELOPMENT POTENTIALTECHNICAL FIELD

[0001] The specification relates to a prediction system generally, and more particularly to a multimodal automated valuation system that integrates with a geographic information system (GIS).BACKGROUND

[0002] Artificial intelligence (Al) refers to technologies that enable computing devices to simulate human intelligence and problem-solving capabilities. Al relies on programs that are trained on sets of data to recognize certain patterns and make decisions or determine outputs without further human intervention. In other words, artificial intelligence utilizes models that apply algorithms to relevant inputs to yield an output.

[0003] Machine Learning (ML) is a subset of Al that may be used to build Al-driven applications. Machine learning is a general umbrella term for techniques and tools that help computing devices learn and adapt on their own. Machine learning algorithms help Al models learn without being explicitly programmed to perform a desired action. By learning a pattern from sample inputs, the machine learning algorithm makes predictions and performs tasks solely based on the learned pattern and not a traditional predefined program instruction.

[0004] Deep Learning is a subset of machine learning that typically utilizes a large volume of data and complex algorithms to train a model. Machine learning can significantly reduce the time, effort, and cost it takes to find solutions to problems, particularly in areas where designing and applying explicit algorithms is either very costly or infeasible.

[0005] One relatively familiar machine learning application is a spam filter for email applications that uses an algorithm to identify incoming junk email. More sophisticated ML applications include automated valuation models (AVMs) that use Al technologies and machine learning algorithms to analyze extensive data and generate property valuations. Many appraisers use AVMs to value residential properties. By eliminating human bias andsubjectivity, AVMs can provide objective property valuations that help estimate market prices. However, existing AVMs rely only on limited, often textual data for their valuations.

[0006] Accordingly, although basic simple approaches of utilizing Al and machine learning for valuations are known, significant improvements in one or more of accuracy, efficiency, automation, and cost are desired.SUMMARY OF THE DISCLOSURE

[0007] In accordance with an aspect of the present disclosure, there is a system for automatically determining an optimal use and valuation of a resource, the system comprising: (i) a multimodal input unit configured to receive a plurality of heterogeneous inputs corresponding to the resource, including: structured metadata from listings for sale of the resource; narrative text descriptions of the resource; images of the resource; historical sales data for the resource; zoning and planning data integrated into a geographic information system (GIS); and neighbourhood-level data describing one or more of: adjacent parcels, zoning patterns, and infrastructure; (ii) a machine learning engine trained on multimodal and temporal datasets to output a valuation data output; and (iii) an optimizer configured to simulate and evaluate a plurality of usage plans for the resource, using the machine learning engine, and select an optimal valuation and an optimal usage scenario.

[0008] The system may utilize a training dataset for the machine learning engine. Such training dataset may include one or more of: structured features from at least one large language model (LLM); resource metadata; vectorized image data; historical price data; and planning data integrated with a geographic information system (GIS) data.

[0009] The resource may be a real estate property. The vectorized image data may be obtained from one or more of an aerial image of the real estate property, an interior image from a multiple listings service (MLS) database and one or more exterior images of the real estate property.

[0010] The machine learning engine may further include a deep learning time-series architecture selected from the group consisting of: Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), Gated Recurrent Units (GRUs), and Deep Belief Networks (DBNs), trained to capture temporal dependencies in multimodal input data.

[0011] Other technical advantages may become readily apparent to one of ordinary skill in the art after review of the following figures and description.BRIEF DESCRIPTIONS OF THE DRAWINGS

[0012] For a better understanding of the embodiments described herein and to show more clearly how the embodiments may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which:

[0013] FIG. 1 is a simplified schematic block diagram of a system in accordance with one embodiment;

[0014] FIG. 2 is a schematic block diagram representation of training processes for the Al engine depicted in FIG. 1;

[0015] FIG. 3 is a simplified block diagram of various hardware components forming part of the valuation system depicted in FIG. 1; and

[0016] FIG. 4 is a flowchart that summarizes several key steps involved in the process of preparing data to generate feature vectors and training of the Al engine depicted in FIG. 1 with the feature vectors.

[0017] Unless otherwise specifically noted, articles depicted in the drawings are not necessarily drawn to scale.DETAILED DESCRIPTION

[0018] The present disclosure describes embodiments of a system comprising an advanced system architecture that implements an automated valuation model (AVM) to assess the market value of real estate properties through a multimodal approach, utilizing a variety of data inputs.

[0019] The present disclosure also describes other embodiments of systems and methods that can predict an optimal use of a resource such as land, subject to various constraints related to the location, zoning laws, nearby amenities, bodies of water, and the like as will be described below.

[0020] In the present disclosure, a system and method for a multimodal, spatiotemporal automated valuation model (AVM) to assess the market value and optimal use of real property is disclosed. The disclosed embodiments leverage heterogeneous multimodal inputs, including structured property metadata such as multiple listings service (MLS) data, number of bedrooms, square footage, age, parking); narrative property descriptions; digital images (interior, exterior, aerial); historical sales records; zoning bylaws; official and secondary planning data; and geographic information data (GlS)-integrated planning layers capturing geospatial proximity to infrastructure, amenities, and environmental constraints.

[0021] An embodiment of a system according to the present disclosure includes multimodal input unit, a machine learning engine and an optimizer. In one specific embodiment, the system employs large language models (LLMs) to extract structured features from unstructured text and image data and integrates these features with geospatial planning data. Unlike conventional AVMs, the system incorporates neighbourhood-level influences by modeling the impact of adjacent parcels, surrounding zoning patterns, and nearby development activity on the subject property.

[0022] In some embodiments, a development potential index (DPI) is factored into an optimizer block of the system. The DPI is a quantified measure of buildability and redevelopment potential, derived from zoning permissions, policy constraints, infrastructure proximity, and neighbourhood dynamics.

[0023] The machine learning engine further includes deep learning temporal architectures such as Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), Gated Recurrent Units (GRUs), and Deep Belief Networks (DBNs), to capture historical patterns across multimodal inputs, enabling prediction of both current market valuations and future appreciation scenarios under various planning and policy changes.

[0024] The optimizer evaluates a plurality of hypothetical land-use and development scenarios, each incorporating both parcel-level and neighbourhood-level variables, to identify the optimal usage plan and projected valuation trajectory.

[0025] Some embodiments of the disclosed system provide technical advantages by: (i) unifying multimodal, spatio-temporal datasets into a valuation framework; (ii) simulating development outcomes under zoning and infrastructure constraints; (iii) forecasting longterm value shifts via temporal deep learning; and (iv) integrating DPI to guide optimal landuse planning.

[0026] In this disclosure, the expression "real estate property", as used in the present disclosure, is intended to include any parcel of land including vacant lands and structures that are permanently attached to the land such as houses, residential or commercial buildings, fences, shops, malls, warehouses, parking lots, and the like.

[0027] In this disclosure, directional terms such as "top," "bottom," "upwards," "downwards," "vertically," and "laterally" are used in the following description for the purpose of providing relative reference only, and are not intended to suggest any limitations on how any article is to be positioned during use, or to be mounted in an assembly or relative to an environment. The use of the word "a" or "an" when used herein in conjunction with the term "comprising" may mean "one," but it is also consistent with the meaning of "one or more," "at least one" and "one or more than one." Any element expressed in the singular form also encompasses its plural form. Any element expressed in the plural form also encompasses its singular form. The term "plurality" as used herein means more than one, for example, two or more, three or more, four or more, and the like.

[0028] In this disclosure, the terms "comprising", "having", "including", and "containing", and grammatical variations thereof, are inclusive or open-ended and do not exclude additional, un-recited elements and / or method steps. The term "consisting essentially of" when used herein in connection with a composition, use or method, denotes that additional elements, method steps or both additional elements and method steps may be present, but that these additions do not materially affect the manner in which the recited composition, method, or use functions. The term "consisting of" when used herein in connection with a composition, use, or method, excludes the presence of additional elements and / or method steps.

[0029] For clarity of illustration, where considered appropriate, reference numerals may be repeated among the Figures to indicate corresponding or analogous elements. In addition, numerous specific details are set forth in order to provide a thorough understanding of the embodiment or embodiments described herein. However, it will be understood by those of ordinary skill in the art that the embodiments described herein may be practiced without these specific details. In other instances, well-known methods, procedures and components have not been described in detail so as not to obscure the embodiments described herein. It should be understood at the outset that, although embodiments are illustrated in the figures and described below, the principles of the present disclosure may be implemented using any number of techniques, whether currently known or not. The present disclosure should in no way be limited to the implementations and techniques illustrated in the drawings and described in this disclosure.

[0030] Various terms used throughout the present description may be read and understood as follows, unless the context indicates otherwise: "or" as used throughout is inclusive, as though written "and / or"; singular articles and pronouns as used throughout include their plural forms, and vice versa; similarly, gendered pronouns include their counterpart pronouns so that pronouns should not be understood as limiting anything described herein to use, implementation, performance, etc. by a single gender; "exemplary" should be understood as "illustrative" or as a non-limiting example, and not necessarily as"preferred" over other embodiments. Further definitions for terms may be set out herein; these may apply to prior and subsequent instances of those terms, as will be understood from a reading of the present description. It will also be noted that the use of the term "a" or "an" will be understood to denote "at least one" in all instances unless explicitly stated otherwise or unless it would be understood to be obvious that it must mean "one".

[0031] Modifications, additions, or omissions may be made to the systems, apparatuses, and methods described herein without departing from the scope of the disclosure. For example, the components of the systems and apparatuses may be integrated or separated. Moreover, the operations of the systems and apparatuses disclosed herein may be performed by more, fewer, or other components and the methods described may include more, fewer, or other steps. Additionally, steps may be performed in any suitable order. As used in this document, "each" refers to each member of a set or each member of a subset of a set.

[0032] Any module, unit, component, server, computer, terminal, engine or device exemplified herein that executes instructions may include or otherwise have access to computer readable media such as storage media, computer storage media, or data storage devices (removable and / or non-removable) such as, for example, magnetic disks, optical disks, or tape. Computer storage media may include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Examples of computer storage media include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information, and which can be accessed by an application, module, or both. Any such computer storage media may be part of the device or accessible or connectable thereto. Further, unless the context clearly indicates otherwise, any processor or controller set out herein may be implemented as a singular processor or as a plurality of processors. The plurality of processors may be arrayedor distributed, and any processing function referred to herein may be carried out by one or by a plurality of processors, even though a single processor may be exemplified. Any method, application or module herein described may be implemented using computer readable / executable instructions that may be stored or otherwise held by such computer readable media and executed by the one or more processors.

[0033] FIG. 1 is a simplified schematic block diagram of an embodiment of a system disclosed herein. As shown, system 10 includes a valuation system 24 that receives input 12 characterizing a resource to be valued such as real estate property or land, to produce an output 28.

[0034] Valuation system 24 uses a trained artificial intelligence (Al) engine 26 to provide output 28. In this embodiment, engine 26 is a deep learning engine. In other embodiments, engine 26 can be a machine learning engine. As will be understood by persons of skill in the art, deep learning is a subset of machine learning.

[0035] A best land-use optimizer 27 uses trained engine 26 to determine the most optimal land use for a specified property. Best land-use optimizer 27 generates numerous hypothetical land use scenarios for a selected property, including different severance scenarios and various building parameters such as the number of bedrooms, square footage, and number of parking spaces.

[0036] The trained Al engine 26 of system 10 is used to predict the property value for each scenario. In this manner, and through the trained Al engine, the system 10 will identify the best land use scenario for a specific property, by maximizing the estimated value.

[0037] In this embodiment, the output 28 can be a single scalar value such as a property value 30 denominated in a particular currency. In other embodiments, a representation 32 of the best use scenario may also be provided as output.

[0038] As noted above, although simple automated valuation models (AVMs) that use Al technologies and machine learning algorithms to analyze extensive data and generateproperty valuations are known, such prior art models rely only on limited, often textual data.

[0039] In contrast, embodiments of the system disclosed in the present disclosure use multimodal input that utilize many additional types of heterogeneous input including text description 14, resource metadata 18, images 16, historical sales data 20 for the resource (e.g., real estate or land) and merged geographic information data (GIS) data 22 that includes one or more of planning data, zoning data, and other information to produce vastly improved valuation for the resource.

[0040] FIG. 2 is a schematic block diagram representation of one embodiment of a training process for the Al engine 26 of valuation system 24. A multimodal training input 34 provides training data to a feature engineering module 40. Some of the data from multimodal training input 34 may be processed or preprocessed by other modules.

[0041] As will be appreciated by persons of skill in the art, training data from heterogenous data sources are converted into formats suitable for use by the Al engine 26 being trained in a process sometimes called featurization. Such formats are called "feature vectors." Such conversion is accomplished by feature engineering module 40.

[0042] Some of the data in multimodal training input 34 is usable directly by feature engineering module 40 whereas others may require pre-processing. Accordingly, a large language model (LLM) 36 processes the text descriptions to convert the text description data into a format usable by feature engineering module 40. LLMs are large, general- purpose language models that can be pre-trained and thereafter used for specific application such as the valuation system 24. In the depicted embodiment, LLM 36 is GPT-4o which is a large-scale, multimodal model which can accept image and text inputs and produce text outputs, provided by Open Al. In this embodiment, valuation system 24 does not fine-tune the LLM 36 as current flagship LLMs such as GPT-4 and GPT-4o are good enough to be used without further fine-tuning. In alternate embodiments, further fine-tuning may be employed.

[0043] As may be appreciated, the context language used to describe real estate properties is not full of jargon or highly technical. This contrasts with medical or engineering applications where heavy jargon is often employed. The generic nature of the context language used in describing real estate properties thus permits the use of pre-built models including the above-noted LLMs without further tuning. In other embodiments however, additional fine-tuning may be used.

[0044] Merging module 38 integrates GIS data 42 with planning data 23, to provide GIS integrated planning data in a format suitable for use by the feature engineering module 40. GIS data 42 may contain many layers with different purposes. GIS merging module 38 populates a connection between planning data 23 with the actual characteristics of the property, in order to obtain valuable features for the property to assess. Most of this task is done by calculating special kinds of intersections which describe the state of the property regarding the requirements of the policy.

[0045] The planning data 23, originating from the zoning bylaw based on an official plan, secondary plan and zoning data, is translated into specific zoning requirements. This zoning information is then digitized into various GIS layers of GIS data 42 that detail land uses and building regulations, such as setbacks, height limits, land use designations, floor area ratios, building coverage etc. These multiple GIS layers collectively form what is known as GIS integrated planning data.

[0046] As contemplated in this embodiment, multimodal training input 34, such as resource metadata 18 and images 16, historical price data 20, are already usable by feature engineering module 40 and thus do not require preprocessing or the use of pre-trained large language models prior to being used by the feature engineering module 40. In other embodiments, the sources of data multimodal training input 34 may not be directly usable by feature engineering module 40 and thus may require preprocessing or use of pre-trained large language models prior to being used by the feature engineering module 40.

[0047] The multimodal training input 34 to Al engine 26 includes: narrative property descriptions, structured property metadata, images or visual data, planning data, and GISData. As illustrated in FIG. 2, multimodal training input 34 includes text descriptions denoted as 14-1,..., 14-n (individually and collectively "text descriptions 14"), images 16- l,...,16-n (individually and collectively "images 16"), structured resource metadata 18-1, ... 18-n (individually and collectively "resource metadata 18"), historical price data 20-1,... 20- n (individually and collectively "historical price data 20"), and planning data 23.

[0048] Resource metadata 18 includes structure metadata that characterizes core property or land characteristics such as the number of bedrooms, bathrooms, total square footage, location, age of the building, number of parking spots and the like. Structured resource metadata may normally be accessible trough a regional multiple listing system (MLS).

[0049] Text descriptions 14 are narrative property descriptions that provide qualitative descriptions that detail a property's features and amenities that may be distinctive. As contemplated in this embodiment, text descriptions include descriptions of elements such as a well-designed floor plan, southern exposure, a swimming pool, an open concept living area that may be conducive to entertaining and family life and the like. In other embodiments, text descriptions may include other descriptions.

[0050] Images 16 include visual data in the form of digital images or photographs of the property, and may include both interior, exterior, and aerial images obtained from private sources or third-party sources like Google Earth™. Images 16 assist in evaluating the resource's aesthetic and functional qualities. For example, information such as the amount of vegetation, degree of privacy, driveway, roof type and condition, certain neighborhood amenities can be extracted from aerial and exterior images. Interior images which include listing images (from MLS databases) may be analyzed to assess the interior condition of the property such as appliances and build quality.

[0051] Planning data 23 refers to data from sources that have bearing on the effect of planning policies on the value of the resource (e.g., land or building). Planning data thus includes information related to the effect of zoning decisions, official plans, secondary plan designations, and specific planning policies and the like. Zoning bylaw affecting the locationof land or property, potential rezoning opportunities and likelihood, future development projects, and forthcoming infrastructure improvements will be taken into account during the resource valuation process.

[0052] GIS data 42 provides analytical perspectives on the limitations set by zoning rules, which are in turn determined by proximity and inclusion standards. GIS data 42 supplies geospatial details about nearby infrastructure and facilities. For example, valuation of a property is impacted by its position in or near a heritage site or a conservation area. Likewise, a property's proximity to parks, bodies of water, and natural areas influences its market price, highlighting how environmental and regulatory factors spatially affect real estate valuations.

[0053] FIG. 3 depicts a simplified block diagram of various hardware elements of a system that includes a remote Al platform 53 and a client computer device 25, forming part of valuation system 24. In the depicted embodiment, a remote Al platform is the Microsoft Azure® Al Infrastructure provided by Microsoft Corporation.

[0054] The diagram in FIG. 3 depicts a simplified schematic block diagram for ease of illustration. Hardware components of the remote Al platform 53 include central processing units (CPUs) including CPU 49a and CPU 49b. As shown CPU 49a has associated graphical processing units (GPUs) in the form of GPU 45a, GPU 45b, and GPU 45c. CPU 49b similarly has associated GPUs 45d, 45e, and 45f. Each one of GPUs 45a, 45b, 45c, 45d, 45e, 45f may hereinafter be referred to, individually and collectively, as "GPUs 45". Each one of GPUs 45a, 45b, 45c, 45d, 45e, 45f has a corresponding associated memory 47a, 47b, 47c, 47d, 47e, 47f respectively as shown and storage (not specifically illustrated). The GPUs may be among the many GPUs provided NVidia Corporation.

[0055] In Al infrastructure platforms, GPUs 45 are typically organized into nodes, racks, or clusters (a group of connected nodes) within data centers and can often be accessed through virtualization in which access to the GPU acceleration is abstracted away and efficiently shared. The remote Al platform 53 relies mainly on GPUs to accelerate trainingperformance as GPUs 45 have simpler cores that are specialized for parallel numerical computations (compared to CPUs).

[0056] Client software application running on device 25 accesses the power of these GPUs residing at data centers through virtualization, via network access (typically the internet), as cloud instances.

[0057] Computer device 25 which may be a laptop, workstation, or smartphone, is used to access Al services from remote Al platform 53. As shown, the computer device 25 has a number of physical components including a processor 44 which may be the form of a central processing unit ("CPU"), random access memory ("RAM") 48, an input / output ("I / O") interface 52, a network interface 56, and non-volatile storage 60. A high-speed interface circuit 64 enables processor 44 to communicate with the other components. Processor 44 executes processor executable instructions in the form of at least an operating system, and one or more applications including the modules that make up valuation system 24 and Al engine 26 depicted in FIG. 1. RAM 48 provides relatively responsive volatile storage to processor 44. I / O interface 52 allows input to be received from one or more peripheral devices, such as a keyboard, a mouse, etc., and outputs information to output devices, such as a display. Network interface 56 permits wired or wireless communication with other computing devices over computer networks such as the Internet. Non-volatile storage 60 stores the operating system and programs, including computer-executable instructions for implementing the software implemented portions of training and evaluation module 62 depicted in FIG. 1 and associated code, data structures and objects.

[0058] During operation of validation system 24, appropriate services and virtual instances are instantiated in the remote Al platform 53 by computer device 25 and used.

[0059] FIG. 4 depicts a flowchart 400 that summarizes several key steps involved in the process of preparing training data, and actual training of the Al engine 26 with the feature vectors. The illustrated sequence of steps is only one example provided for ease of illustration, and the steps shown in such example need not always be performed in thedepicted order. It would be understood by persons of skill in the art that some orders may be swapped or concurrently carried out without affecting the end result.

[0060] Unlike conventional computer vision approaches where machine learning (ML) is used to process images, a cutting-edge LLM technology (i.e., GPT-4o) is utilized. GPT-4o excels in at understanding and discussing images. GPT-4o is especially adept at vision and audio comprehension compared to existing LLMs.

[0061] Referring to step 402, textual data that supplies a narrative property description that undergoes feature engineering to extract selected elements that include upgrades or specific appliance types, which are then structured for further analysis. LLM 36 is utilized to preprocess text inputs that provide narrative property descriptions. As noted earlier, the Al services including LLM 36 are remotely accessed from remote Al platform 53, in the form of Microsoft Azure® Al Infrastructure.

[0062] Narrative property description is sent as an input to GPT-4o which then provides pre-defined feature parameters to be included as input in the valuation system 24.

[0063] Referring to step 404, digital images of a property or land are provided to the LLM 36 (e.g., GPT-4o) for feature extraction and a set of tags are generated which describe the image. The generated features will be included in the training process.

[0064] At step 405, metadata preprocessing and preparation are carried out. Planning data 23 undergoes a preprocessing in which sources of planning information are transformed into a GIS-compatible format by merging module 38.

[0065] Referring to step 406, each attribute is mapped and linked to a specific geographical section of land. Planning data 23, is initially mostly unstructured, and unsuitable for querying. Planning data documents are thus populated, and a query-able repository of planning data is created for use with model training and evaluation module 62.

[0066] The conversion ensures that planning regulations, such as land use designations and zoning laws, are spatially defined and accurately represented across the designated areas, facilitating more precise analysis and application in spatial planning contexts.

[0067] Referring to step 408, GIS data preprocessing is performed by one or more of the merging module 38 and feature engineering module 40. Geometric operations, such as calculating boundary intersections, buffers, distances, and area calculations, are used to output structured data that includes metrics representative of the land's geospatial value. For example, proximity to a railroad, a school, or public transportation hubs affect the computed value. The resulting data is formatted as set of feature vectors suitable for input into the Al engine 26.

[0068] Referring to step 410, integration and analysis is carried out by feature engineering module 40 to combine engineered features with structured metadata and analyzed by Al engine 26, which further assesses image or visual data to determine factors such as the property's condition and layout.

[0069] Referring to step 412, feature vectors are ready to be used for training the Al engine 26.

[0070] Referring to steps 414 and 416, training is carried out on extensive historical real estate data, and the parameters are updated as needed, to enable Al engine 26 to identify how different features influence resource values and predict future trends and appreciation rates.

[0071] According to an embodiment of system 10, the output 28 generated by the valuation system 24 is the estimated current market value of the property, factoring in both intrinsic characteristics and historical market data.

[0072] In other embodiments, other types of output can be realized. For example, scenario planning may be carried out by inputting different type of properties on a given piece of land. Varying the number bedrooms or building in appropriate hooks to easily convert the type of property in anticipation of future zoning changes, transportation facilities and the like allows for planning of the best use of a piece of land.

[0073] Advantageously, embodiments of the system described in the present disclosure, by taking the planning data into account, can provide information regarding land usage that was previously unattainable. For example, a parcel of land may be 10 acres in area of which only 2 acres can be developed for residential or commercial purposes. In contrast, another parcel of land nearby may be only 5 acres in area but the valuation system 24 may accurately determine that the smaller parcel of land is more valuable since all of the land may be developed for commercial or residential use. The use of planning data integrated with GIS allows this advantageous feature of embodiments of the system disclosed herein where other conventional AVMs would miss the effect of planning data.

[0074] Among other advantageous innovations of the present disclosure are neighbourhood influence modeling, temporal deep learning architectures, and a development potential index (DPI) which are defined below.

[0075] Neighbourhood influence modeling: refers to capturing patterns from neighbouring parcels and surrounding real estate that influence the subject parcel's valuation.

[0076] Temporal deep learning architectures: refers to incorporation of RNNs, LSTMs, GRUs, and Deep Belief Networks to capture historical multimodal trends and forecast future valuations.

[0077] Development Potential Index (DPI): refers to inclusion of a Development Potential Index as one of the multimodal features factored into the optimizer for valuation and best-use determination.

[0078] Consideration of planning data and its effects on the value of a given property is a particularly advantageous feature of embodiments of the present disclosure. Additionally, using LLMs to extract data from images instead of ML models confers advantages. Integrating LLM results with AVM ML models is a further advantage of embodiments of the present disclosure.Neighbourhood & Spatial Extensions

[0079] As noted above, in accordance with an aspect, a system for automatically determining an optimal use and valuation of a resource, includes: a multimodal input unit configured to receive a plurality of heterogeneous inputs corresponding to the resource, including at least: structured metadata from listings for sale of the resource; narrative property descriptions of the resource; images of the resource; historical sales data for the resource; zoning and planning data integrated into a geographic information system (GIS); and neighbourhood-level data describing one or more of: adjacent parcels, zoning patterns, and infrastructure; a machine learning engine trained on multimodal and temporal datasets to output valuation data output; and an optimizer configured to simulate and evaluate a plurality of usage plans for the resource using the machine learning engine and select an optimal valuation and an optimal usage scenario.

[0080] In accordance with another aspect, the machine learning engine further incorporates features derived from neighbouring parcels of land and surrounding real estate properties, including zoning, infrastructure, and development attributes that influence the valuation of the subject resource.

[0081] In accordance with another aspect, the optimizer evaluates the effect of neighbourhood-level patterns comprising land-use designations, historical transactions, infrastructure development, and adjacent property characteristics on the valuation of the resource.Temporal Extensions

[0082] In accordance with another aspect, the machine learning engine further comprises a deep learning time-series architecture selected from the group consisting of Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), Gated Recurrent Units (GRUs), and Deep Belief Networks (DBNs), trained to capture temporal dependencies in multimodal input data.

[0083] In accordance with another aspect, the deep learning time-series architecture or the temporal architecture is configured to analyze historical multimodal data including property metadata, planning data, images, and geospatial datasets to predict future valuations and trends for the resource.

[0084] In accordance with another aspect, the temporal architecture is further configured to evaluate a plurality of hypothetical land-use scenarios across time, thereby forecasting changes in value under anticipated zoning changes, infrastructure developments, or market shifts.Development Potential Index (DPI)In accordance with another aspect the optimizer such as optimizer 27 of FIG. 1 further incorporates a Development Potential Index (DPI). The DPI includes a quantified measure of buildability and redevelopment capacity of the resource, derived from multimodal inputs including planning data, zoning regulations, infrastructure proximity, and neighbourhood trends. In accordance with another aspect the DPI is dynamically adjusted based on projected policy changes, infrastructure upgrades, or neighbourhood redevelopment activity. In accordance with yet another aspect wherein the DPI may be applied to other applications beyond valuation such as city planning, lending and underwriting, insurance risk modeling, and portfolio optimization for real estate investment trusts (REITs).

[0085] Although specific advantages have been enumerated above, various embodiments may include some, none, or all of the enumerated advantages.

[0086] Persons skilled in the art will appreciate that there are yet more alternative implementations and modifications possible, and that the above examples are only illustrations of one or more implementations. The scope, therefore, is only to be limited by the claims appended hereto and any amendments made thereto.

Claims

CLAIMSWhat is claimed is:

1. A system for automatically determining an optimal use and valuation of a resource, the system comprising:(a) a multimodal input unit configured to receive a plurality of heterogeneous inputs corresponding to the resource, comprising :(i) structured metadata from listings for sale of the resource;(ii) narrative text descriptions of the resource;(iii) images of the resource;(iv) historical sales data for the resource;(v) zoning and planning data integrated into a geographic information system (GIS); and(vi) neighbourhood-level data describing one or more of: adjacent parcels, zoning patterns, and infrastructure;(b) a machine learning engine trained on multimodal and temporal datasets to output a valuation data output; and(c) an optimizer configured to simulate and evaluate a plurality of usage plans for the resource, using the machine learning engine, and select an optimal valuation and an optimal usage scenario.

2. The system of claim 1, wherein the valuation data output comprises one or more of: a market price, a valuation forecast, and a recommended usage plan, for the resource.

3. The system of claim 1, wherein the resource is a real estate property.

4. The system of claim 1, wherein the resource is land.

5. The system of claim 1, wherein the machine learning engine utilizes at least one large language model (LLM), the at least one LLM receiving and vectorizing said images and said narrative text descriptions into structured features.

6. The system of claim 5, wherein a training dataset for the machine learning engine comprises one or more of:(a) said structured features from the at least one large language model (LLM);(b) resource metadata;(c) vectorized image data;(d) historical price data; and(e) planning data integrated with a geographic information system (GIS) data.

7. The system of claim 6, wherein the resource is a real estate property, and the resource metadata comprises structured data from a multiple listings service (MLS) database.

8. The system of claim 6, wherein the resource is a real estate property, and wherein the vectorized image data is obtained from one or more of an aerial image of the real estate property, an interior image from a multiple listings service (MLS) database and one or more exterior images of the real estate property.

9. The system of claim 3, wherein the valuation data output is a predicted market value of the real estate property.

10. The system of claim 3, wherein the valuation data output is a range of predicted market values of the real estate property.

11. The system of claim 1, wherein the machine learning engine further comprises features derived from characteristics of neighbouring parcels of land and surrounding real estate properties, said characteristics comprising zoning, infrastructure, and development attributes that influence the valuation of the resource.

12. The system of claim 11, wherein the optimizer evaluates effects of neighbourhoodlevel patterns comprising land-use designations, historical transactions, infrastructure development, and adjacent property characteristics on the valuation of the resource.

13. The system of claim 3, wherein the machine learning engine further comprises a deep learning time-series architecture selected from the group consisting of: Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), Gated Recurrent Units (GRUs), and Deep Belief Networks (DBNs), trained to capture temporal dependencies in multimodal input data.

14. The system of claim 13, wherein the deep learning time-series architecture is configured to analyze historical multimodal data comprising property metadata, planning data, images, and geospatial datasets, to predict future valuations and trends for the resource.

15. The system of claim 13, wherein the deep learning time-series architecture is further configured to evaluate a plurality of hypothetical land-use scenarios across time, thereby forecasting changes in value under anticipated zoning changes, infrastructure developments, or market shifts.

16. The system of claim 1, wherein the optimizer further incorporates a Development Potential Index (DPI), the DPI comprising a quantified measure of buildability and redevelopment capacity of the resource, derived from multimodal inputs including planning data, zoning regulations, infrastructure proximity, and neighbourhood trends.

17. The system of claim 16, wherein the DPI is dynamically adjusted based on one or more of: projected policy changes, infrastructure upgrades, and neighbourhood redevelopment activity.

18. The system of claim 16, wherein the DPI is applied activities comprising one or more of: city planning, lending and underwriting, insurance risk modeling, and portfolio optimization for real estate investment trusts (REITs).

19. The system of claim 3, wherein the valuation data output is a predicted market value of the real estate property.

20. The system of claim 3, wherein the valuation data output is a range of predicted market values of the real estate property.

21. The system of claim 2, wherein the recommended usage plan is selected from the plurality of usage plans for the resource.

22. The system of claim 3, the structured metadata comprises data from a multiple listings service (MLS) database.

Images