Search

CN-122019878-A - Information recommendation method, device, system and storage medium

CN122019878ACN 122019878 ACN122019878 ACN 122019878ACN-122019878-A

Abstract

The invention discloses an information recommendation method, an information recommendation device, an information recommendation system and a storage medium, and relates to the technical field of computer software and data processing. The method includes obtaining an ordered sequence of elements Resource upper limit In (1) Constrained scan Based on cost function Computing element resource consumption and accumulating to determine last non-overrun position of prefix Identifying overrun elements, elements that enter near the upper bound interval, and/or importance scores that are not below a threshold Elements forming a candidate post-inclusion element set Generating a state abstract ; Will Input decision module (rule-based decision strategy and/or a learning model) determines prefix end position And post-contain fragment sets Based on And Generating a recommended subsequence And output. The method can generate the information recommendation result under the constraint of multidimensional resources, and is suitable for the segment information recommendation scenes such as texts, dialogs, logs, multimedia and the like.

Inventors

  • Ye Guoda

Assignees

  • 叶国达

Dates

Publication Date
20260512
Application Date
20260201

Claims (8)

  1. 1. An information recommendation method, characterized by comprising the following steps: Step one, obtaining an ordered element sequence Resource upper limit ; Step two, at the upper limit of the resource Constrained to said ordered sequence of elements Performing a scan process based on a cost function Computing element resource consumption and accumulating to determine the last non-overrun position of the prefix Construction of candidate post-containing element sets And generate a state abstract Wherein the scanning process includes the steps of Computing heuristic resource consumption And is based on And (3) with After the first occurrence of the resource overrun in the scanning process, continuing to scan the ordered element sequence to supplement the candidate post-containing element set But without enlarging the scanning prefix ; Step three, abstracting the state Input decision module based on the state abstract Determining prefix end position And then contains fragment sets ; Step four, based on the prefix end position The post-containing fragment set Generating a recommended subsequence And outputs, wherein the post contains a set of fragments Represented as one or more sequences of elements from the ordered sequence of elements Sets of extracted fragments 。
  2. 2. The information recommendation method according to claim 1, wherein the resource upper limit Is a multidimensional resource upper limit vector, and the scanning process comprises the steps of aiming at elements Computing resource consumption vectors And accumulating the resource consumption vectors at the current prefix Forming heuristic resource consumption vector on the basis When there is a resource dimension Satisfy the following requirements And determining that the resource exceeds the limit, and determining the last non-exceeding position of the prefix Corresponding prefix resource consumption vector 。
  3. 3. The information recommendation method according to claim 1 or 2, wherein said candidate post-inclusion element set At least comprises one corresponding element of the following: Overrun candidate element, existing resource dimension Satisfy the following requirements Is an element of (2); near the upper limit candidate element, on the premise of not exceeding the limit, the resource dimension exists Satisfy the following requirements Wherein The buffer parameters are preset; High importance candidate element, importance score Satisfy the following requirements Wherein , And presetting an importance threshold value.
  4. 4. An information recommendation method according to any one of claims 1 to 3, further comprising including a dedicated resource budget after acquisition And in generating the recommended subsequence And respectively checking the total resource consumption and the later-contained partial resource consumption: effective budget for post-definition contained parts As dimension-wise minimum When (when) When not provided or certain dimensions are not configured, the corresponding dimension is considered to impose no additional constraints; Overall resource consumption vector for recommended subsequences Checking to meet In (1) When provided, the pair post contains partial resource consumption vector Checking to meet 。
  5. 5. The information recommendation method according to any one of claims 1 to 4, wherein said decision module determines said prefix end position using one of The post-containing fragment set Rule-based decision strategies, learning model-based decision strategies, or a combination of both.
  6. 6. An information recommendation device configured to perform the information recommendation method according to any one of claims 1 to 5, the device comprising: A receiving module for receiving the ordered element sequence Upper limit of resources Optionally post-inclusion dedicated resource budgets And/or policy configuration; A scan modeling module for determining a resource upper limit Constrained to said ordered sequence of elements Scanning to determine the last non-overrun position of the prefix And constructing a candidate post-containing element set And generates a status abstract Wherein after resource overrun occurs for the first time in the scanning process, the scan modeling module continues to scan the ordered element sequence to supplement the candidate post-inclusion element set But without enlarging the scanning prefix ; A decision module for abstracting based on the state Determining prefix end position And then contains fragment sets ; Generating an output module for ending a position based on the prefix The post-containing fragment set Generating and outputting a recommended subsequence 。
  7. 7. An information recommendation system, comprising the information recommendation device as claimed in claim 6 and at least one client or service caller; The client or service caller is configured to send an ordered sequence of elements to the information recommendation device Upper limit of resources And transmitting an optional post-inclusion dedicated resource budget And/or policy configuration; The information recommendation device is configured to execute the information recommendation method according to any one of claims 1 to 5 and return a recommendation sub-sequence to the client or service caller 。
  8. 8. A computer readable storage medium having stored thereon computer program instructions, which, when executed by a processor, are adapted to carry out the information recommendation method according to any of claims 1 to 5.

Description

Information recommendation method, device, system and storage medium Technical Field The invention relates to the technical field of computer software and data processing, in particular to an information recommendation method for carrying out candidate screening and combination on an ordered element sequence and generating a recommendation sub-sequence based on a strategy under the condition of multidimensional resource constraint, and an information recommendation device, an information recommendation system and a computer readable storage medium for realizing the method. Background In many computer application scenarios, it is desirable to select a portion of the elements from a number of candidate elements in a sequence of ordered data of indeterminate length, present the result to the user as a recommendation (i.e., a recommendation subsequence) or as input to a subsequent process, and cause the recommendation to satisfy a certain upper resource constraint. For example: 1. Text processing and dialog system recommendation In text summarization, dialog system prompt word construction (e.g., context construction of large language models), it is necessary to defineAnd selecting partial contents from ordered elements such as historical conversations, system prompts, knowledge segments and the like within the limit of numbers or characters as recommendation results, and inputting the recommendation results into the model. The conventional method mostly adopts simple prefix truncation, namely, elements are spliced sequentially from the beginning of the sequence, and once the upper limit is exceeded, the elements are stopped, and additional consideration is not carried out on the subsequent elements, so that information (such as the latest user request, error prompt or conclusion sentence) which is positioned at the tail but is very critical cannot be recommended to the model for use. 2. Log segment recommendation and transmission In the log collection and reporting scenario, in order to save network bandwidth and storage resources, it is often necessary to select a part of the log in the log sequence as a recommended segment for uploading or displaying. Existing schemes typically employ "keep recentlyBar log "or" pre-reservationThe simple strategy of the log is difficult to consider the background context and the section of the abnormal information of the log under the condition of limited resources, so that a better information coverage effect cannot be achieved when the log is recommended to operation and maintenance personnel or an analysis system. 3. Multimedia data segment recommendation In the video or audio summary and preview scene, a plurality of fragments are selected from the complete video or audio sequence as recommended content to form the preview or summary in a limited play time. The traditional method mainly comprises uniform extraction or simple prefix interception, and is difficult to flexibly recommend according to key event positions, user attention points and the like, so that recommendation fragments cannot fully embody the key points of media contents. To sum up, the prior art has the following problems in the "recommendation under ordered data sequence" scenario: Multiple kinds of resource constraints (such as length, element number, processing time, memory occupation, etc.) cannot be considered simultaneously due to the adoption of single dimension (such as length) constraint, so that the recommendation result is difficult to finely control the resource consumption; recommendation strategies are usually simple prefix selection or simple tail reservation, and lack recommendation mechanisms which are intelligently contained based on overrun points or key positions; a unified framework containing element sets after candidate is not established, and one or more local fragment combinations cannot be flexibly selected to be recommended results; The comprehensive utilization of the characteristics such as element importance and category is lacking, and the information value of the recommendation result cannot be maximized on the premise of limited resources; the decision logic is mainly fixed hard coding rules, and is difficult to configure according to a service scene or optimize through data driving, so that the recommendation strategy is not intelligent enough and has insufficient expandability. Therefore, in order to solve the above-mentioned problems, it is necessary to design a more flexible information recommendation method and apparatus, which can perform prefix selection and candidate post-inclusion decision on an ordered element sequence under a multidimensional resource constraint condition, and generate a recommendation subsequence including prefixes and optional post-inclusion fragments in combination with a configurable or leachable strategy, thereby improving flexibility, information value and adaptability of a recommendation result. Disclosure of Invention The invention provi