Improved system-user interaction relating to advice

Abstract

The current invention concerns an improved system-user interaction relating to advice, specifically an allocation proposition which assigns certain values to a plurality of physical and optionally virtual entities.

Description

[0001]

[0002] Technical field

[0003] The invention pertains to the technical field of computer network processing techniques on a large data stream, whereby certain user preferences are received, and an allocation proposition over a plurality of physical and optionally virtual entities is created based on said data stream, the user preferences and other factors.

[0004] Background

[0005] The applicant noticed that there is a lack of computer-implemented systems which are capable of distributing finite quantities over a plurality of physical and / or virtual entities, whereby the distribution takes into account information from a plurality of information sources, user-prompted preferences (which comprise among others a valuation or user rating of said information sources) and factors which at least define geographical information and application field information for each of the entities. Said proposed distribution is compared to one or more comparative compositions which are based on the user preferences and the factors, and offered to the user based on said comparison.

[0006] US 1,3935,198 describes a system that governs a supply chain dynamically, by analysis of outside information sources, such as news channels, regarding possibly influencing vents, and estimating an impact (logistical and / or financial) on the supply chain. However, this is a rigid system that is focused entirely on a supply chain, and cannot be rapidly modified to suit other needs, or specific user preferences.

[0007] Other similar systems, as described in US 7,680,719, are known in the art, but are lacking either in user interaction (in that a user who does not have administrative power to change the inner workings of the system, cannot provide personalized preferences), or in that the used information sources are not properly screened, or in that there is no ‘practical’ filter, which, for instance, would weed out geographically impossible or highly unlikely suggestions.

[0008] The invention thereto aims to address at least some of the problems mentioned. Summary of the invention

[0009] In a first aspect, the present invention provides a method according to claim 1.

[0010] In a second aspect, the present invention provides a system according to claim 14.

[0011] In a further aspect, the present invention provides a use according to claim 15.

[0012] Further preferred embodiment and their advantages are described in the detailed description and the claims.

[0013] Detailed description of the invention

[0014] As used herein, the following terms have the following meanings:

[0015] “A”, “an”, and “the” as used herein refers to both singular and plural referents unless the context clearly dictates otherwise. By way of example, “a compartment” refers to one or more than one compartment.

[0016] “Comprise”, “comprising”, and“comprises” and“comprised of” as used herein are synonymous with “include”, “including”, “includes” or “contain”, “containing”, “contains” and are inclusive or open-ended terms that specifies the presence of what follows e.g. component and do not exclude or preclude the presence of additional, non-recited components, features, element, members, steps, known in the art or disclosed therein.

[0017] The recitation of numerical ranges by endpoints includes all numbers and fractions subsumed within that range, as well as the recited endpoints.

[0018] The term information source may relate to any data channel, data stream, data set, or data subscription service providing content in a file-based, update-based, synchronization-based and / or message-based format, whereby each information source not providing data in pure messages may be considered as equivalent to message based, whereby updated portions of data may relate to one or more new messages. The messages may be provided on demand by requesting it from storage, or may be sent in real time to the user, e.g., as an e-mail, a push message notification, an in-app message, a dedicated web interface, a web dashboard, or a news item in a feed. The m essages m ay be delivered in the context of a subscription and / or an account with the provider of an inform ation source. Unless otherwise defined, all terms used in disclosing the invention , including techn ical and scientific term s, have the m ean ing as com m on ly u nderstood by one of ordinary skill in the art to wh ich this invention belongs. By m eans of fu rther gu idance, term definitions are included to better appreciate the teach ing of the present invention .

[0019] The present invention in a first aspect concerns a com puter- im plem ented m ethod for providing a system -generated executable notification for a user, based on interaction between a user and said system , com prising the steps of :

[0020] a. prom pting the user for providing a value for a plurality of user preferences to the system ;

[0021] b. repeatedly receiving, by a server, a plurality of m essages originating from a plurality of inform ation sources, said plu rality of m essages being directly and / or indirectly related to one or more predefined physical entities and one or m ore predefined virtual entities, whereby the received m essages are stored as context data;

[0022] c. assigning at least one factor of each of at least two factor sets to each of the physical entities and to each of the virtual entities, whereby a first of said at least two factor sets relates to a geographical valuation of the physical and virtual entities, and a second of said at least two factor sets relates to an application field of the physical and virtual entities;

[0023] d. creating an allocation proposition , by allocating values to a plurality of the physical and virtual entities, whereby the aggregated allocated value of the physical entities is non-zero, and whereby the aggregated allocated value of the physical and virtual entities is equal to a user-predefined value, said allocation proposition com prising the physical and virtual entities with a non-zero allocated value, and each of the associated non-zero allocated values;

[0024] e. providing an executable notification to the user, whereby said executable notification com prises the allocation proposition ;

[0025] characterized in that at least one of the user preferences define a user valuation of each of the inform ation sou rces;

[0026] whereby the allocated values are calcu lated based on the values of the user preferences, the factors and the m essages of the inform ation sources, whereby the allocation proposition is com pared to a com parative com position comprising a plurality of physical and virtual entities with a non-zero allocated value, whereby the aggregated non-zero allocated value of the entities of the comparative composition equal the user-predefined value, whereby the allocated values of the comparative composition are based on the values for the user preferences and the factors, and whereby the executable notification is provided based on the comparison of the allocation proposition with the com parative com position.

[0027] The above defined m ethod offers an advantageous solution for entity m anagement in num erous fields. Given the versatility of the m ethod defined, said entities m ay vary strongly in their definition , ranging from various tangible tradable goods, such as foodstuffs, cars, raw m aterials, general supplies, coupons, art, property, livestock and others, to more elusive tradable goods such as stocks, equity, fixed income, cash , real estate, precious goods, investm ent fund and others.

[0028] A distribution of values for such entities may for instance be assigning a location for storage, a proprietor, a price, a storage cost, storage characteristics, transfer of said entities, etc. An optim ized distribution of such param eters / values is strongly dependent on many agents, some explicit, others im plicit or indirect.

[0029] A first of said agents can be user preferences, which often has a strong determ inative quality in the distribution, and can act as a guide for the other agents to run an optim ization , especially as the user preferences - in this case - reflect on the use of at least som e of the other agents.

[0030] A second agent is the plurality of inform ation sources. As m entioned, the user preferences amongst others define a user valuation of said information sources, which guides the allocation process, in that the valuation determ ines the‘weight’ that each inform ation source has in the allocation process. One user m ay prefer certain inform ation sources to other (for instance Twitter feed over official Dow Jones news) , and subvaluations may exist that further specify the weight for certain subsets of inform ation of som e information sources.

[0031] Thirdly, an agent may be a so-called ‘factor’ from a factor set. Said factors are assigned to each of the entities, and define a certain characteristic of the entity. At least one of said factors defines geographical inform ation of the entity, and at least one factor defines an application field of the entity. The geographical inform ation typically is a country, region or continent (but could be a city, a state, or a group of several of one or more of the above) , that is representative for the entity. For instance, in the case of physical objects such as raw m aterials, this could be their point of origin (production area or the likes of such) , or their actual physical location . I n the case of virtual objects, this may represent the location of their proprietor or the location of a physical entity they are related to (such as in the case of stock, the location or region of the company that issues said stock). As for the application field, this may reflect the nature of the entity (for instance as defined above: stock, raw material, property, food, medical device, etc.), or more generally, the technical field to which it belongs: healthcare, technology, other. It is found that such factors often will have a strong influence on an optimal allocation. For instance, in the case of distribution of entities such as food, the geographical location will be crucial, as will be its ‘category’ (meat, dairy, grain, conservable, etc.) for optimal allocation. The same applies to, for instance, the construction of a portfolio of investments, where it is crucial that the entities there (whether physical or virtual) are defined by such factors, as location may influence security of physical goods, and thus their reliability / value, or may influence volatility in case of stock.

[0032] Furthermore, the factors of the entities define a base that allows the computer system to establish links to certain messages that impact said entities, which simplifies the calculation for the allocated values of the entities.

[0033] In these cases, user-defined constraints must often be placed on the allocated values, whether it be due to a limited available cargo space (weight and / or volume), limited refrigeration means, manpower, and / or even budgetary constrictions. As such, these are set as a maximal value for the aggregate allocated values to the entities.

[0034] In the end, the system generates an allocation proposition which basically comprises a listing of all of the entities with a non-zero allocated value, and the associated allocated value of said entities. This can be provided to and further analyzed by a (qualified) person and authorized, with or without changes or corrections (for instance, should the person have knowledge of information that was not in the system yet, or simply has certain objections due to previous experiences).

[0035] Flowever, before the allocation proposition is provided to the user, it is compared to one or more other allocation propositions to verify the validity / efficiency. These ‘known’ comparative compositions may be taken from historical compositions in the past (both old system-generated compositions, user-generated propositions, and outside information, so-called benchmark distributions). In order to satisfy the requirement to be provided to the user, the allocation proposition must ‘outperform’ the comparative composition based on certain criteria.

[0036] Finally, the allocation proposition is provided as an actionable notification, which can be understood that it can be executed (possibly modifiable by the user as well) to affect a (technical) effect in reality. In this document, the term “value” or “allocated value” can be understood to represent certain information regarding to an entity. These values may be an amount of said entity which is proposed to be ‘used’, as in: transported, purchased, stored,

[0037] None of the prior art systems allows such an intricate interaction between a user and a system in order to generate allocation propositions. For instance, systems will generally bar the user from imparting a valuation on their input information. Flowever, it should be understood that human experience can expertly guide the system in correctly valuing certain information sources, dependent on the subject (for instance, when distributing emergency supplies to combat a humanitarian crisis, a user may impart a greater value on ‘quick’ social media sources, such as Twitter, possibly with a further bias to believe certain specific trustworthy accounts against others, over classic sources, such as newspapers, that cannot deliver urgent content at the same speed). Oppositely, more volatile news sources (typically self-published, such as Twitter, Facebook, or self-published web content, etc.) can be rated lower than the traditional news sources, or other more regulated channels, when time is less an issue.

[0038] In a preferred embodiment, said user preferences comprise at least a regularization type preference and a regularization level preference;

[0039] preferably whereby said regularization type preference relates to either regularization without variable selection or to regularization with variable selection;

[0040] said regularization without variable selection more preferably relating to Tikhonov regularization or ridge-type regularization;

[0041] said regularization with variable selection more preferably relating to lasso regularization;

[0042] and preferably whereby said regularization level preference relates to a value, more preferably a real value, indicative of the intensity of said regularization.

[0043] Flereby, “regularization” refers to any mechanism that evaluates the messages of the information sources in view of the uncertainty and / or significance of the information source providing the messages that impact the output of the evaluation. This addresses the problem that despite messages being generated constantly, with a high level of “noise”, i.e. a high uncertainty, a large part of the messages hardly (if at all) contributes to the eventual values generated by the evaluation. Regularization thus reflects a level of distrust with respect to the outputs generated by the evaluation, based on mathematical notions of stability. Regularization can hence be understood as a three step approach:

[0044] define some notion of normality for each information source, typically an absence of messages, or a“zero signal”;

[0045] model the stream of messages from every information source as a sum of content and noise, or“real signal + noise”;

[0046] remove the noise by pushing the output of the evaluation, or “the total signal”, closer to what is normal.

[0047] In one variant, the regularization boils down to automatically scaling the impact of incoming messages according to their uncertainty, while retaining all messages from all sources, and preferably updating a reference value of the evaluation for each arrival. The triggering of a notification may then relate to messages from any information source, whereby regularization may suppress some of the anomalies or outliers. Hereby, the user may preferably control the intensity of said regularization via said regularization level preference. In a corresponding and / or related preferred embodiment, said regularization type preference relates to regularization without variable selection; said regularization without variable selection preferably relating to Tikhonov regularization or ridge-type regularization; wherein said regularization level preference relates to a value, preferably a real value, indicative of the intensity of said regularization.

[0048] In another variant, the regularization corresponds to automatically suppressing information sources, for some period in time, for which messages are either deemed too uncertain, or of too little significance, while scaling the other messages according to their uncertainty. The allocation proposition may then be based only on messages from only the information sources that were not suppressed, whereby regularization may further scale the impact of those messages that are not suppressed. Hereby, the user may preferably control the intensity of said regularization via said regularization level preference. In a corresponding and / or related preferred embodiment, said regularization type preference relates to regularization with variable selection; said regularization with variable selection preferably relating to lasso regularization; wherein said regularization level preference relates to a value, preferably a real value, indicative of the intensity of said regularization. With regularization, the user is further provided with a useful tool to assess the value of various information sources, which greatly supports the working of the allocation of values to the entities.

[0049] In a preferred embodiment, the values allocated to the entities of the allocation proposition are calculated furthermore based on at least one user-selected optimization model, said optimization model selected from a list comprising at least Black-Litterman model, Markowitz model and minimum variance model.

[0050] In a preferred embodiment, the information sources comprise a direct context information source, a structured peripheral information source, an unstructured peripheral information source and a conjectural direct context information source, whereby data of the direct context information source is enriched via feature engineering, and preferably whereby data of the conjectural direct context information source and / or of the structured peripheral information source is enriched via feature engineering.

[0051] In a preferred embodiment, said plurality of information sources comprises at least one unstructured peripheral information source;

[0052] wherein at least one message originating from said unstructured peripheral information source comprises a raw text string;

[0053] comprising a step of comparing said message to said context data using natural language processing of said raw text string for assessing a relevance of said message with respect to said context data;

[0054] whereby the allocated values are calculated based furthermore on said assessment.

[0055] In a further preferred embodiment, messages from the unstructured peripheral information source are processed before use in calculated the allocated values, whereby said processing comprises a structuring of said messages of the unstructured peripheral information source via a self-learning artificial intelligence, whereby said artificial intelligence operates under user-tunable hyperparameters.

[0056] In a preferred embodiment, the physical and virtual entities are each attributed a predictability rating based on at least historical information of said physical and virtual entities in the context data, whereby the allocated values of the allocation proposition are furthermore calculated based on said predictability ratings. Knowledge of the ‘predictability’ of certain entities can aid in forecasting further developments, and can be useful in creating a certain balance or security in the total allocation proposition.

[0057] In a preferred embodiment, the prompted user preferences comprise one or more constraints, said constraints comprising at least a user-defined minimal predictability rating of the physical and / or virtual entities with a non-zero allocated value in the allocation proposition, and / or a user-defined minimal aggregate predictability rating of the physical and virtual entities with a non-zero allocated value in the allocation proposition.

[0058] As mentioned, when ignoring the predictability rating of entities, it is possible that a very risky allocation proposition is created, which may be efficient or advantageous, but only in the supposition that everything goes according to plan. Even when a moderate stance is taken on the future effects of each entity, there will still be a strong difference in the potential variation of the effects for each entity (for instance disastrous harvests, production problems, etc. will impact certain entities stronger), which is not reflected in a forecast. Therefore, depending on the situation (urgency, location ...), a user may allow a higher or lower predictability rating of the allocation proposition.

[0059] In a preferred embodiment, the prompted user preferences comprise one or more constraints for creating the allocation proposition, said constraints comprising at least a maximal number and / or minimal number of physical and / or virtual entities with a non-zero allocated value in the allocation proposition, and / or a maximal and / or minimal value for the allocated values of the physical and / or virtual entities in the allocation proposition.

[0060] In a preferred embodiment, the user preferences comprise a historical allocation composition which comprises one or more of the physical and / or virtual entities with non-zero historical values,

[0061] whereby an objective cost function is defined in the system based on the context data which is adapted for calculating a rebalancing cost for converting the historical allocation composition into the allocation proposition.

[0062] The historical allocation composition comprises a‘current’, or‘latest’ (or expected) allocation composition, representing the last known values for the entities (or at least for the entities which have a non-zero allocated value in either the allocation proposition and / or in the historical allocation composition. In some cases, changing the values will bring forth a certain cost, be it administrative, governmental, or practical, such as certain taxes, transport costs, etc. Note that this cost need not necessarily be‘monetary’, but can be a cost in time to enact a certain change, or a cost in personnel (manpower), or even in space (volume / weight). In order to present an allocation proposition, these costs must be taken into account by the user, as they can impact the advantage of the presented allocation proposition (for instance, a too high time cost may negate all other advantages of a proposition).

[0063] In a preferred embodiment, the prompted user preferences comprise at least one constraint for creating the allocation proposition, whereby said constraint is a maximal value for the rebalancing cost for converting the historical allocation composition into the allocation proposition.

[0064] In a preferred embodiment, the context data is furthermore processed via ensemble learning. By combining multiple learning algorithms, the predictive performance of the entire process is improved.

[0065] In a preferred embodiment, the user preferences comprise at least one constraint on the context data on which the calculation of the allocated values is based, whereby said constraint is dependent on at least one of the following: period in which the message comprising said context data is generated, user valuation of the information source of the message comprising said context data. This allows the user to put a focus on information from a certain period, for instance, in order to simplify the evaluation process.

[0066] In a second aspect, the invention pertains to a computer system for generating an executable notification for a user, based on interaction between a user and said system, comprising:

[0067] a. a server, the server comprising a processor, tangible non-volatile memory, program code present on said memory for instructing said processor, connection means for connecting to a device of said user and to one or more remote servers;

[0068] b. said device of said user, the user device comprising a processor, tangible non-volatile memory, program code present on said memory for instructing said processor, a screen for displaying information to said user, preferably input means for receiving a user input m eans from a user, connection m eans for connecting to said server via said com puter network;

[0069] c. said one or more remote servers, each of said remote servers relating to at least one of said plurality of inform ation sources, each of said one or more remote servers com prising connection means for connecting to said server via said computer network;

[0070] d. at least one com puter-readable medium , the at least one com puter- readable medium accessible to said server and comprising a database, said database com prising context data and user preference data, the at least one computer-readable medium preferably com prised in said server;

[0071] said computer system configured for generating said executable notification for said user, whereby generating said executable notification com prises the following steps:

[0072] prompting the user for providing a value for a plurality of user preferences to the system ;

[0073] repeatedly receiving, by a server, a plurality of messages originating from a plurality of information sources, said plurality of m essages being directly and / or indirectly related to one or more predefined physical entities and one or more predefined virtual entities, whereby the received m essages are stored as context data;

[0074] assigning at least one factor of each of at least two factor sets to each of the physical entities and to each of the virtual entities, whereby a first of said at least two factor sets relates to a geographical valuation of the physical and virtual entities, and a second of said at least two factor sets relates to an application field of the physical and virtual entities;

[0075] creating an allocation proposition , by allocating values to a plurality of the physical and virtual entities, whereby the aggregated allocated value of the physical entities is non-zero, and whereby the aggregated allocated value of the physical and virtual entities is equal to a user-predefined value, said allocation proposition comprising the physical and virtual entities with a non-zero allocated value, and each of the associated non-zero allocated values;

[0076] providing an executable notification to the user, whereby said executable notification com prises the allocation proposition ;

[0077] whereby at least one of the user preferences defines a user valuation of each of the information sources;

[0078] whereby the allocated values are calculated based on the values of the user preferences, the factors and the m essages of the information sources, whereby the allocation proposition is com pared to a com parative com position comprising a plurality of physical and virtual entities with a non-zero allocated value, whereby the aggregated non-zero allocated value of the entities of the comparative com position equal the user-predefined value, whereby the allocated values of the comparative composition are based on the values for the user preferences and the factors.

[0079] I n a third aspect, the invention pertains to the use of the m ethod according to the invention in a system according to the invention .

[0080] The invention is further described by the following non-lim iting exam ples which further illustrate the invention, and are not intended to, nor should they be interpreted to, lim it the scope of the invention .

[0081] Exam pies

[0082] Example 1 : Physical entity collection

[0083] I n this exam ple, the invention relates to the monitoring by a user of a physical object collection , said physical object collection comprising a plurality of physical goods or physical objects stored or located at a plurality of locations. The physical objects m ay for instance be precious goods with intrinsic value such as gold, diamonds, (oldtimer) cars, wine, whiskey, art paintings, musical instrum ents, or jewelry, but may also relate to precious goods of which the value is related to the right that is conferred to the owner, such as bearer bonds. I n yet another example the physical objects may relate to real estate, e.g. an object m ay correspond to a house, a building, an apartm ent, an apartm ent block, a garage, a garage block. I n a preferred example embodiment, each of the objects corresponds to a physical entity record stored in a database and vice versa. I n one embodiment, the physical objects may relate to perishable goods which may suffer damage due to an incident relating to excessive tem peratures or high moisture levels, as is e.g. the case for cars, wine, whiskey, art paintings, musical instrum ents, etc. Hereby, a first allocation proposition m ay relate to m axim izing the total number of objects across locations that remains intact over a given tim e period, e.g. , one year. For such an allocation proposition , in view of m easurem ents and predictions for different locations available from different information sources, a notification m ay be generated whenever any of the locations is likely to suffer an incident based on a m essage originating from one of the inform ation sources. However, a second allocation proposition m ay relate to m axim izing the chance that at least one object rem ains intact over the given tim e period of, e.g. , 1 year. This may for instance relate to hard disk drives or flash drives com prising highly confidential inform ation which is duplicated across multiple locations, whereby the hard disk drives or flash drives may or m ay not be connected to a com puter network depending on the level of confidentiality required, and whereby the data on the drives m ay or m ay not be encrypted. Such highly confidential information may for instance relate to passwords, credentials or cryptocurrency assets. For such an allocation proposition , predictions or m easurem ents relating to a single location likely lack relevance and hence do not require a notification , unless a large majority of the objects is stored at that single location. Whether the latter is the case may be indicated by a strong alteration of the allocation proposition’s output in case an indicator of an incident at that single location is detected. I n various related exam ple embodim ents, one inform ation source may be an alert service offering alerts relating to the integrity and safety of locations and / or buildings, such as the detection of an alarm or incident at an object location. Another inform ation source m ay relate to a raw text service providing news m essages. Flereby, the detection of said the name of said location and / or the nam e of a building relating to the location may indicate that the m essage has relevance in view of the allocation proposition .

[0084] An advantage of the invention lies in the control by the user of what is relevant and what is not, and to what extent notifications are required. Regularization enables selection of inform ation sources which are m eaningful to the end user. For instance, a news m essage from a single information source regarding a fire at a location m ay not be significant and m ay be deemed an outlier based on regularization, whereas multiple messages with highly sim ilar content m ay indicate high relevance, and hence trigger a notification . Flereby, the strength of regularization allows to set a threshold depending on user preference. Moreover, the user m ay indicate whether he prefers regularization without variable selection or with variable selection . Overall, the user is relieved of the burden of m anually going through all m essages with potential relevance for the physical object collection. With regularization, the user is further provided with a useful tool to assess the value of various information sources. The invention can hence be of value in deciding which subscriptions to inform ation sources (paying or not) should be kept and which may be removed. Example 2: Portfolio of tradable goods

[0085] I n this example, the context data relates to all inform ation relating to a portfolio com prising a plurality of tradable goods of which at least two relate to a physical entity. The context data com prises a plurality of data records including at least two physical entity records, each of the physical entity records relating to a physical entity. The physical entity may be any physical investm ent product, for instance gold, diamonds, (oldtim er) cars, wine, whiskey, art paintings, collector’s items, m usical instrum ents, jewelry, bearer bonds, a house, a building, an apartment, an apartm ent block, a garage, a garage block. Beyond physical entity records, the portfolio may or may not comprise data records which do relate to entities that are not physical entities. Each of the data records is a structured set of entries for a number of attributes. One attribute may be the type of entity, e.g. stocks, equity, fixed incom e, cash , real estate, precious goods, investment fund, etc. Another attribute may indicate an amount of said entity. Yet another attribute may relate to a starting date or an ending date of a com m itm ent related to said entity. I n analogy with Exam ple 1 , another attribute may be a location of said entity, specified, e.g. , in terms of a street address or a set of GPS coordinates. Another attribute m ay allow for a natural language description of the entity, which may be taken into account through natural language processing.

[0086] The allocation proposition relates to a technical aim for said portfolio. The form ulation of the allocation proposition relates to a training of a plurality of parameters on the history of each of the information sources. The allocation proposition may or may not further relate to an application of the Black-Litterman optim ization. The information sources include pre-structured data com prising records comprising entries for a plurality of fields, as well as unstructured data com prising raw text. The pre-structured data provides indicators for news intensity, news sentim ent and news readership, both at company level and at aggregate level. The pre-structured data m ay relate to an index. The user is further provided with possibilities of changing the inform ation sources upon which a sentim ent score is built, to cater to evolutions in social m edia use, evolutions in form al press. Hereby, sentim ent is typically m easured relative to some historical standard, but the relevant standard can be modified to emphasize belief in certain opinion m akers. The process flow can be changed as well, with e.g. the date update being daily or more frequent. The automated process hence feeds the hum an process, with added value for the portfolios being monitored. Particularly, the invention provides a com prehensive alternative to purely manual processing, provides a partial alternative to sub decisions, and provides inform ation generated in the process. Big shifts in forecasts can trigger a notification with explanation which is sent to users, e.g. portfolio m anagers, who can then rebalance their portfolios. The invention further allows to identify whether certain entities and / or the related financial instruments are being overvalued. Furthermore, the user may be warned of high volum es of negative news building up around a certain entity. Moreover, when e.g. the stock of som e com pany has plum m eted, an indirect link with several entities m ay be established based on linked com panies and sectors that m ight be affected.

[0087] The way in which a notification m ay be generated for a user is further analogous to Example 1 . Also in this exam ple, the user is assisted in the burden of going through all m essages with potential relevance for the portfolio. With regularization , the user is further provided with a useful tool to assess the value of various inform ation sources. The invention can hence be of value in deciding which subscriptions to information sources (paying or not) should be kept and which m ay be removed

[0088] Example 3: Crisis supply distribution

[0089] I n this exam ple, the context data relates to information regarding a regional crisis situation, such as a fam ine, outbreak of disease, natural disaster, etc. The information sources in this case be traditional media (tv, newspaper, online coverage) , social m edia (Twitter, Facebook, I nstagram , Tencent, Weibo, Youtube) , and others. The entities in this case m ay relate to food, m edicinal supplies, infrastructural supplies (tents, ...) , financial support and others. The user m ay then provide certain values in response to prompts of the system , amongst which a valuation of the information sources. I n these cases, unstructured information sources can be very relevant as they quickly identify dangerous situations on a local level (for instance certain villages where an outbreak has erupted) , while larger m edia coverage will m iss these or refuse to report without further confirmation . The user may thus give a greater valuation to such sources. Other information that will be im parted are geographical details, and the nature of the crisis.

[0090] The system will have access to a database on a large amount of entities with information on said entities, for instance grain supplies in surrounding areas (countries, regions, cities, etc.) , m edicinal supplies as well as m edical personnel from certain organizations ( Doctors without Borders, ...) , transport capacity (X number of cargo planes with Y capacity in country Z, ...) . Note that said database will be updated via the influx of information from the information sources.

[0091] Based on this, the system can generate, on time-basis, a proposal for allocating resources to address a crisis. Said proposal will allocate values to a plurality of entities, which values can indicate whether or not these will be used (for instance, if certain supplies cannot be provided in tim e from certain locations, they will not be used) , how much of them will be used, and how they will be provided.

[0092] I t is supposed that the present invention is not restricted to any form of realization described previously and that som e modifications can be added to the presented exam ple of fabrication without reappraisal of the appended claims. For example, the present invention has been described referring to distributions of goods, com modities and others, both physical and virtual, but it is clear that the invention can be used for other applications.

Claims

CLAI MS1 . Computer-im plemented method for providing a system -generated executable notification for a user, based on interaction between a user and said system , com prising the steps of :a. prompting the user for providing a value for a plurality of user preferences to the system ;b. repeatedly receiving, by a server, a plurality of m essages originating from a plurality of information sources, said plurality of m essages being directly and / or indirectly related to one or more predefined physical entities and one or more predefined virtual entities, whereby the received m essages are stored as context data;c. assigning at least one factor of each of at least two factor sets to each of the physical entities and to each of the virtual entities, whereby a first of said at least two factor sets relates to a geographical valuation of the physical and virtual entities, and a second of said at least two factor sets relates to an application field of the physical and virtual entities;d. creating an allocation proposition, by allocating values to a plurality of the physical and virtual entities, whereby the aggregated allocated value of the physical entities is non-zero, and whereby the aggregated allocated value of the physical and virtual entities is equal to a user-predefined value, said allocation proposition comprising the physical and virtual entities with a non-zero allocated value, and each of the associated non-zero allocated values;e. providing an executable notification to the user, whereby said executable notification comprises the allocation proposition ;characterized in t hat at least one of the user preferences define a user valuation of each of the information sources;whereby the allocated values are calculated based on the values of the user preferences, the factors and the messages of the inform ation sources, whereby the allocation proposition is com pared to a comparative composition comprising a plurality of physical and virtual entities with a non-zero allocated value, whereby the aggregated non-zero allocated value of the entities of the com parative composition equal the user-predefined value, whereby the allocated values of the com parative composition are based on the values for the user preferences and the factors, and wherebythe executable notification is provided based on the comparison of the allocation proposition with the comparative com position .

2. Computer-implemented m ethod according to the preceding claim 1 , characterized in t hat said user preferences comprise at least a regularization type preference and a regularization level preference;preferably whereby said regularization type preference relates to either regularization without variable selection or to regularization with variable selection ;said regularization without variable selection more preferably relating to Tikhonov regularization or ridge-type regularization ;said regularization with variable selection more preferably relating to lasso regularization ;and preferably whereby said regularization level preference relates to a value, more preferably a real value, indicative of the intensity of said regularization .

3. Com puter-implem ented m ethod according to any of the preceding claims 1 or 2, characterized in t hat the values allocated to the entities of the allocation proposition are calculated furthermore based on at least one user-selected optim ization model, said optim ization model selected from a list com prising at least Black-Litterm an model, Markowitz model and m inim um variance model.

4. Com puter-implem ented method according to any of the preceding claims 1 to 3, characterized in t hat the inform ation sources comprise a direct context inform ation source, a structured peripheral information source, an unstructured peripheral information source and a conjectural direct context inform ation source, whereby data of the direct context information source is enriched via feature engineering, and preferably whereby data of the conjectural direct context information source and / or of the structured peripheral inform ation source is enriched via feature engineering.

5. Computer-implemented m ethod according to any one of the preceding claims 1 to 4, characterized in t hat said plurality of information sources comprises at least one unstructured peripheral inform ation source;wherein at least one m essage originating from said unstructured peripheral inform ation source com prises a raw text string;comprising a step of com paring said m essage to said context data using natural language processing of said raw text string for assessing a relevance of said m essage with respect to said context data;whereby the allocated values are calculated based furthermore on said assessm ent.

6. Com puter-implem ented method according to the preceding claim 4 or 5, characterized in t hat m essages from the unstructured peripheral inform ation source are processed before use in calculated the allocated values, whereby said processing com prises a structuring of said messages of the unstructured peripheral inform ation source via a self-learning artificial intelligence, whereby said artificial intelligence operates under user-tunable hyperparam eters.

7. Com puter-implem ented m ethod according to any one of the preceding claims 1 to 6, characterized in t hat the physical and virtual entities are each attributed a predictability rating based on at least historical inform ation of said physical and virtual entities in the context data, whereby the allocated values of the allocation proposition are furthermore calculated based on said predictability ratings.

8. Com puter-implem ented method according to any one of the preceding claims 1 to 7, characterized in t hat the prompted user preferences comprise one or more constraints, said constraints comprising at least a user-defined m inimal predictability rating of the physical and / or virtual entities with a non-zero allocated value in the allocation proposition , and / or a user-defined m inim al aggregate predictability rating of the physical and virtual entities with a non-zero allocated value in the allocation proposition .

9. Com puter-implem ented m ethod according to any of the preceding claims 1 to 8, characterized in t hat the prom pted user preferences comprise one or more constraints for creating the allocation proposition, said constraints comprising at least a m aximal num ber and / or m inim al num ber of physical and / or virtual entities with a non-zero allocated value in the allocation proposition , and / or a maxim al and / or m inimal value for the allocated values of the physical and / or virtual entities in the allocation proposition .

10. Computer-implemented method according to any of the preceding claims 1 to 9, characterized in that the user preferences comprise a historical allocation composition which comprises one or more of the physical and / or virtual entities with non-zero historical values,whereby an objective cost function is defined in the system based on the context data which is adapted for calculating a rebalancing cost for converting the historical allocation composition into the allocation proposition.

11. Computer-implemented method according to the preceding claim 10, characterized in that the prompted user preferences comprise at least one constraint for creating the allocation proposition, whereby said constraint is a maximal value for the rebalancing cost for converting the historical allocation composition into the allocation proposition.

12. Computer-implemented method according to any of the preceding claims 1 to 11, characterized in that the context data is furthermore processed via ensemble learning.

13. Computer-implemented method according to any of the preceding claims 1 to 12, characterized in that the user preferences comprise at least one constraint on the context data on which the calculation of the allocated values is based, whereby said constraint is dependent on at least one of the following: period in which the message comprising said context data is generated, user valuation of the information source of the message comprising said context data.

14. A computer system for generating an executable notification for a user, based on interaction between a user and said system, comprising:a. a server, the server comprising a processor, tangible non-volatile memory, program code present on said memory for instructing said processor, connection means for connecting to a device of said user and to one or more remote servers;b. said device of said user, the user device comprising a processor, tangible non-volatile memory, program code present on said memory for instructing said processor, a screen for displaying information to said user, preferably input means for receiving a userinput means from a user, connection m eans for connecting to said server via said com puter network;c. said one or more remote servers, each of said remote servers relating to at least one of said plurality of inform ation sources, each of said one or more remote servers comprising connection means for connecting to said server via said computer network;d. at least one com puter-readable m edium , the at least one computer- readable m edium accessible to said server and comprising a database, said database com prising context data and user preference data, the at least one computer-readable m edium preferably com prised in said server;said computer system configured for generating said executable notification for said user, whereby generating said executable notification com prises the following steps:prompting the user for providing a value for a plurality of user preferences to the system ;repeatedly receiving, by a server, a plurality of messages originating from a plurality of information sources, said plurality of m essages being directly and / or indirectly related to one or more predefined physical entities and one or more predefined virtual entities, whereby the received m essages are stored as context data;assigning at least one factor of each of at least two factor sets to each of the physical entities and to each of the virtual entities, whereby a first of said at least two factor sets relates to a geographical valuation of the physical and virtual entities, and a second of said at least two factor sets relates to an application field of the physical and virtual entities;creating an allocation proposition, by allocating values to a plurality of the physical and virtual entities, whereby the aggregated allocated value of the physical entities is non-zero, and whereby the aggregated allocated value of the physical and virtual entities is equal to a user- predefined value, said allocation proposition com prising the physical and virtual entities with a non-zero allocated value, and each of the associated non-zero allocated values;providing an executable notification to the user, whereby said executable notification comprises the allocation proposition ;characterized in t hat at least one of the user preferences defines a user valuation of each of the inform ation sources;whereby the allocated values are calculated based on the values of the user preferences, the factors and the messages of the information sources, whereby the allocation proposition is compared to a comparative composition comprising a plurality of physical and virtual entities with a non-zero allocated value, whereby the aggregated non-zero allocated value of the entities of the comparative composition equal the user-predefined value, whereby the allocated values of the comparative composition are based on the values for the user preferences and the factors.

15. Use of the method according to any one of the preceding claims 1 to 13 in the system according to the preceding claim 14.