# frozen_string_literal: true
# Copyright (c) 2017-2018 THL A29 Limited, a Tencent company. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
module TencentCloud
module Nlp
module V20190408
# AnalyzeSentiment请求参数结构体
class AnalyzeSentimentRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分析的文本(仅支持UTF-8格式,不超过200字)。
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# AnalyzeSentiment返回参数结构体
class AnalyzeSentimentResponse < TencentCloud::Common::AbstractModel
# @param Positive: 正面情感概率。
# @type Positive: Float
# @param Neutral: 中性情感概率。
# @type Neutral: Float
# @param Negative: 负面情感概率。
# @type Negative: Float
# @param Sentiment: 情感分类结果:
# positive:正面情感
# negative:负面情感
# neutral:中性、无情感
# @type Sentiment: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Positive, :Neutral, :Negative, :Sentiment, :RequestId
def initialize(positive=nil, neutral=nil, negative=nil, sentiment=nil, requestid=nil)
@Positive = positive
@Neutral = neutral
@Negative = negative
@Sentiment = sentiment
@RequestId = requestid
end
def deserialize(params)
@Positive = params['Positive']
@Neutral = params['Neutral']
@Negative = params['Negative']
@Sentiment = params['Sentiment']
@RequestId = params['RequestId']
end
end
# AutoSummarization请求参数结构体
class AutoSummarizationRequest < TencentCloud::Common::AbstractModel
# @param Text: 待处理的文本(仅支持UTF-8格式,不超过2000字)
# @type Text: String
# @param Length: 指定摘要的长度上限(默认值为200)
# 注:为保证摘要的可读性,最终生成的摘要长度会低于指定的长度上限。
# @type Length: Integer
attr_accessor :Text, :Length
def initialize(text=nil, length=nil)
@Text = text
@Length = length
end
def deserialize(params)
@Text = params['Text']
@Length = params['Length']
end
end
# AutoSummarization返回参数结构体
class AutoSummarizationResponse < TencentCloud::Common::AbstractModel
# @param Summary: 文本摘要结果
# @type Summary: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Summary, :RequestId
def initialize(summary=nil, requestid=nil)
@Summary = summary
@RequestId = requestid
end
def deserialize(params)
@Summary = params['Summary']
@RequestId = params['RequestId']
end
end
# 基础粒度分词和词性标注的结果
class BasicParticiple < TencentCloud::Common::AbstractModel
# @param Word: 基础词。
# @type Word: String
# @param BeginOffset: 基础词在NormalText中的起始位置。
# @type BeginOffset: Integer
# @param Length: 基础词的长度。
# @type Length: Integer
# @param Pos: 词性。
# @type Pos: String
attr_accessor :Word, :BeginOffset, :Length, :Pos
def initialize(word=nil, beginoffset=nil, length=nil, pos=nil)
@Word = word
@BeginOffset = beginoffset
@Length = length
@Pos = pos
end
def deserialize(params)
@Word = params['Word']
@BeginOffset = params['BeginOffset']
@Length = params['Length']
@Pos = params['Pos']
end
end
# 文本纠错结果
class CCIToken < TencentCloud::Common::AbstractModel
# @param Word: 错别字内容
# @type Word: String
# @param BeginOffset: 错别字的起始位置,从0开始
# @type BeginOffset: Integer
# @param CorrectWord: 错别字纠错结果
# @type CorrectWord: String
attr_accessor :Word, :BeginOffset, :CorrectWord
def initialize(word=nil, beginoffset=nil, correctword=nil)
@Word = word
@BeginOffset = beginoffset
@CorrectWord = correctword
end
def deserialize(params)
@Word = params['Word']
@BeginOffset = params['BeginOffset']
@CorrectWord = params['CorrectWord']
end
end
# 分类详细信息
class Category < TencentCloud::Common::AbstractModel
# @param Id: 分类id。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Id: Integer
# @param Label: 分类英文名。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Label: String
# @param Name: 分类中文名。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Name: String
# @param Score: 分类置信度。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Score: Float
attr_accessor :Id, :Label, :Name, :Score
def initialize(id=nil, label=nil, name=nil, score=nil)
@Id = id
@Label = label
@Name = name
@Score = score
end
def deserialize(params)
@Id = params['Id']
@Label = params['Label']
@Name = params['Name']
@Score = params['Score']
end
end
# ChatBot请求参数结构体
class ChatBotRequest < TencentCloud::Common::AbstractModel
# @param Query: 用户请求的query
# @type Query: String
# @param OpenId: 服务的id, 主要用于儿童闲聊接口,比如手Q的openid
# @type OpenId: String
# @param Flag: 0: 通用闲聊, 1:儿童闲聊, 默认是通用闲聊
# @type Flag: Integer
attr_accessor :Query, :OpenId, :Flag
def initialize(query=nil, openid=nil, flag=nil)
@Query = query
@OpenId = openid
@Flag = flag
end
def deserialize(params)
@Query = params['Query']
@OpenId = params['OpenId']
@Flag = params['Flag']
end
end
# ChatBot返回参数结构体
class ChatBotResponse < TencentCloud::Common::AbstractModel
# @param Reply: 闲聊回复
# @type Reply: String
# @param Confidence: 对于当前输出回复的自信度
# @type Confidence: Float
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Reply, :Confidence, :RequestId
def initialize(reply=nil, confidence=nil, requestid=nil)
@Reply = reply
@Confidence = confidence
@RequestId = requestid
end
def deserialize(params)
@Reply = params['Reply']
@Confidence = params['Confidence']
@RequestId = params['RequestId']
end
end
# 文本分类结果
class ClassificationResult < TencentCloud::Common::AbstractModel
# @param FirstClassName: 一级分类名称
# @type FirstClassName: String
# @param SecondClassName: 二级分类名称
# @type SecondClassName: String
# @param FirstClassProbability: 一级分类概率
# @type FirstClassProbability: Float
# @param SecondClassProbability: 二级分类概率
# @type SecondClassProbability: Float
# @param ThirdClassName: 三级分类名称,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type ThirdClassName: String
# @param ThirdClassProbability: 三级分类概率,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type ThirdClassProbability: Float
# @param FourthClassName: 四级分类名称,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type FourthClassName: String
# @param FourthClassProbability: 四级分类概率,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type FourthClassProbability: Float
# @param FifthClassName: 五级分类名称,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type FifthClassName: String
# @param FifthClassProbability: 五级分类概率,仅有当新闻领域五分类可能出现,详情见文本分类文档
# @type FifthClassProbability: Float
attr_accessor :FirstClassName, :SecondClassName, :FirstClassProbability, :SecondClassProbability, :ThirdClassName, :ThirdClassProbability, :FourthClassName, :FourthClassProbability, :FifthClassName, :FifthClassProbability
def initialize(firstclassname=nil, secondclassname=nil, firstclassprobability=nil, secondclassprobability=nil, thirdclassname=nil, thirdclassprobability=nil, fourthclassname=nil, fourthclassprobability=nil, fifthclassname=nil, fifthclassprobability=nil)
@FirstClassName = firstclassname
@SecondClassName = secondclassname
@FirstClassProbability = firstclassprobability
@SecondClassProbability = secondclassprobability
@ThirdClassName = thirdclassname
@ThirdClassProbability = thirdclassprobability
@FourthClassName = fourthclassname
@FourthClassProbability = fourthclassprobability
@FifthClassName = fifthclassname
@FifthClassProbability = fifthclassprobability
end
def deserialize(params)
@FirstClassName = params['FirstClassName']
@SecondClassName = params['SecondClassName']
@FirstClassProbability = params['FirstClassProbability']
@SecondClassProbability = params['SecondClassProbability']
@ThirdClassName = params['ThirdClassName']
@ThirdClassProbability = params['ThirdClassProbability']
@FourthClassName = params['FourthClassName']
@FourthClassProbability = params['FourthClassProbability']
@FifthClassName = params['FifthClassName']
@FifthClassProbability = params['FifthClassProbability']
end
end
# ClassifyContent请求参数结构体
class ClassifyContentRequest < TencentCloud::Common::AbstractModel
# @param Title: 待分类的文章的标题(仅支持UTF-8格式,不超过100字符)。
# @type Title: String
# @param Content: 待分类文章的内容, 每个元素对应一个段落。(仅支持UTF-8格式,文章内容长度总和不超过2000字符)
# @type Content: Array
attr_accessor :Title, :Content
def initialize(title=nil, content=nil)
@Title = title
@Content = content
end
def deserialize(params)
@Title = params['Title']
@Content = params['Content']
end
end
# ClassifyContent返回参数结构体
class ClassifyContentResponse < TencentCloud::Common::AbstractModel
# @param FirstClassification: 一级分类。(请参见附录[三级分类体系表](https://cloud.tencent.com/document/product/271/94286))
# @type FirstClassification: :class:`Tencentcloud::Nlp.v20190408.models.Category`
# @param SecondClassification: 二级分类。(请参见附录[三级分类体系表](https://cloud.tencent.com/document/product/271/94286))
# @type SecondClassification: :class:`Tencentcloud::Nlp.v20190408.models.Category`
# @param ThirdClassification: 三级分类。(请参见附录[三级分类体系表](https://cloud.tencent.com/document/product/271/94286))
# 注意:此字段可能返回 null,表示取不到有效值。
# @type ThirdClassification: :class:`Tencentcloud::Nlp.v20190408.models.Category`
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :FirstClassification, :SecondClassification, :ThirdClassification, :RequestId
def initialize(firstclassification=nil, secondclassification=nil, thirdclassification=nil, requestid=nil)
@FirstClassification = firstclassification
@SecondClassification = secondclassification
@ThirdClassification = thirdclassification
@RequestId = requestid
end
def deserialize(params)
unless params['FirstClassification'].nil?
@FirstClassification = Category.new
@FirstClassification.deserialize(params['FirstClassification'])
end
unless params['SecondClassification'].nil?
@SecondClassification = Category.new
@SecondClassification.deserialize(params['SecondClassification'])
end
unless params['ThirdClassification'].nil?
@ThirdClassification = Category.new
@ThirdClassification.deserialize(params['ThirdClassification'])
end
@RequestId = params['RequestId']
end
end
# ComposeCouplet请求参数结构体
class ComposeCoupletRequest < TencentCloud::Common::AbstractModel
# @param Text: 生成对联的关键词。长度需>=2,当长度>2时,自动截取前两个字作为关键字。内容需为常用汉字(不含有数字、英文、韩语、日语、符号等等其他)。
# @type Text: String
# @param TargetType: 返回的文本结果为繁体还是简体。0:简体;1:繁体。默认为0。
# @type TargetType: Integer
attr_accessor :Text, :TargetType
def initialize(text=nil, targettype=nil)
@Text = text
@TargetType = targettype
end
def deserialize(params)
@Text = params['Text']
@TargetType = params['TargetType']
end
end
# ComposeCouplet返回参数结构体
class ComposeCoupletResponse < TencentCloud::Common::AbstractModel
# @param TopScroll: 横批。
# @type TopScroll: String
# @param Content: 上联与下联。
# @type Content: Array
# @param RandomCause: 当对联随机生成时,展示随机生成原因。
# @type RandomCause: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :TopScroll, :Content, :RandomCause, :RequestId
def initialize(topscroll=nil, content=nil, randomcause=nil, requestid=nil)
@TopScroll = topscroll
@Content = content
@RandomCause = randomcause
@RequestId = requestid
end
def deserialize(params)
@TopScroll = params['TopScroll']
@Content = params['Content']
@RandomCause = params['RandomCause']
@RequestId = params['RequestId']
end
end
# ComposePoetry请求参数结构体
class ComposePoetryRequest < TencentCloud::Common::AbstractModel
# @param Text: 生成诗词的关键词。
# @type Text: String
# @param PoetryType: 生成诗词的类型。0:藏头或藏身;1:藏头;2:藏身。默认为0。
# @type PoetryType: Integer
# @param Genre: 诗的体裁。0:五言律诗或七言律诗;5:五言律诗;7:七言律诗。默认为0。
# @type Genre: Integer
attr_accessor :Text, :PoetryType, :Genre
def initialize(text=nil, poetrytype=nil, genre=nil)
@Text = text
@PoetryType = poetrytype
@Genre = genre
end
def deserialize(params)
@Text = params['Text']
@PoetryType = params['PoetryType']
@Genre = params['Genre']
end
end
# ComposePoetry返回参数结构体
class ComposePoetryResponse < TencentCloud::Common::AbstractModel
# @param Title: 诗题,即输入的生成诗词的关键词。
# @type Title: String
# @param Content: 诗的内容。
# @type Content: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Title, :Content, :RequestId
def initialize(title=nil, content=nil, requestid=nil)
@Title = title
@Content = content
@RequestId = requestid
end
def deserialize(params)
@Title = params['Title']
@Content = params['Content']
@RequestId = params['RequestId']
end
end
# 复合粒度分词和词性标注的结果。
class CompoundParticiple < TencentCloud::Common::AbstractModel
# @param Word: 基础词。
# @type Word: String
# @param BeginOffset: 基础词在NormalText中的起始位置。
# @type BeginOffset: Integer
# @param Length: 基础词的长度。
# @type Length: Integer
# @param Pos: 词性。
# @type Pos: String
attr_accessor :Word, :BeginOffset, :Length, :Pos
def initialize(word=nil, beginoffset=nil, length=nil, pos=nil)
@Word = word
@BeginOffset = beginoffset
@Length = length
@Pos = pos
end
def deserialize(params)
@Word = params['Word']
@BeginOffset = params['BeginOffset']
@Length = params['Length']
@Pos = params['Pos']
end
end
# 纠错结果列表
class CorrectionItem < TencentCloud::Common::AbstractModel
# @param Order: 纠错句子的序号。
# @type Order: Integer
# @param BeginOffset: 错误的起始位置,从0开始。
# @type BeginOffset: Integer
# @param Len: 错误内容长度。
# @type Len: Integer
# @param Word: 错误内容。
# @type Word: String
# @param CorrectWord: 纠错结果,当为删除类错误时,结果为null。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type CorrectWord: Array
# @param CorrectionType: 纠错类型。0:替换;1:插入;2:删除。
# @type CorrectionType: Integer
# @param Confidence: 纠错信息置信度。0:error;1:warning;error的置信度更高。(仅供参考)
# @type Confidence: Integer
# @param DescriptionZh: 纠错信息中文描述。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type DescriptionZh: String
# @param DescriptionEn: 纠错信息英文描述。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type DescriptionEn: String
attr_accessor :Order, :BeginOffset, :Len, :Word, :CorrectWord, :CorrectionType, :Confidence, :DescriptionZh, :DescriptionEn
def initialize(order=nil, beginoffset=nil, len=nil, word=nil, correctword=nil, correctiontype=nil, confidence=nil, descriptionzh=nil, descriptionen=nil)
@Order = order
@BeginOffset = beginoffset
@Len = len
@Word = word
@CorrectWord = correctword
@CorrectionType = correctiontype
@Confidence = confidence
@DescriptionZh = descriptionzh
@DescriptionEn = descriptionen
end
def deserialize(params)
@Order = params['Order']
@BeginOffset = params['BeginOffset']
@Len = params['Len']
@Word = params['Word']
@CorrectWord = params['CorrectWord']
@CorrectionType = params['CorrectionType']
@Confidence = params['Confidence']
@DescriptionZh = params['DescriptionZh']
@DescriptionEn = params['DescriptionEn']
end
end
# CreateDict请求参数结构体
class CreateDictRequest < TencentCloud::Common::AbstractModel
# @param Name: 自定义词库名称,不超过20字。
# @type Name: String
# @param Description: 自定义词库描述,不超过100字。
# @type Description: String
attr_accessor :Name, :Description
def initialize(name=nil, description=nil)
@Name = name
@Description = description
end
def deserialize(params)
@Name = params['Name']
@Description = params['Description']
end
end
# CreateDict返回参数结构体
class CreateDictResponse < TencentCloud::Common::AbstractModel
# @param DictId: 创建的自定义词库ID。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type DictId: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :DictId, :RequestId
def initialize(dictid=nil, requestid=nil)
@DictId = dictid
@RequestId = requestid
end
def deserialize(params)
@DictId = params['DictId']
@RequestId = params['RequestId']
end
end
# CreateWordItems请求参数结构体
class CreateWordItemsRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param WordItems: 待添加的词条集合。
# @type WordItems: Array
attr_accessor :DictId, :WordItems
def initialize(dictid=nil, worditems=nil)
@DictId = dictid
@WordItems = worditems
end
def deserialize(params)
@DictId = params['DictId']
unless params['WordItems'].nil?
@WordItems = []
params['WordItems'].each do |i|
worditem_tmp = WordItem.new
worditem_tmp.deserialize(i)
@WordItems << worditem_tmp
end
end
end
end
# CreateWordItems返回参数结构体
class CreateWordItemsResponse < TencentCloud::Common::AbstractModel
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :RequestId
def initialize(requestid=nil)
@RequestId = requestid
end
def deserialize(params)
@RequestId = params['RequestId']
end
end
# DeleteDict请求参数结构体
class DeleteDictRequest < TencentCloud::Common::AbstractModel
# @param DictId: 要删除的自定义词库ID。
# @type DictId: String
attr_accessor :DictId
def initialize(dictid=nil)
@DictId = dictid
end
def deserialize(params)
@DictId = params['DictId']
end
end
# DeleteDict返回参数结构体
class DeleteDictResponse < TencentCloud::Common::AbstractModel
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :RequestId
def initialize(requestid=nil)
@RequestId = requestid
end
def deserialize(params)
@RequestId = params['RequestId']
end
end
# DeleteWordItems请求参数结构体
class DeleteWordItemsRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param WordItems: 待删除的词条集合。
# @type WordItems: Array
attr_accessor :DictId, :WordItems
def initialize(dictid=nil, worditems=nil)
@DictId = dictid
@WordItems = worditems
end
def deserialize(params)
@DictId = params['DictId']
unless params['WordItems'].nil?
@WordItems = []
params['WordItems'].each do |i|
worditem_tmp = WordItem.new
worditem_tmp.deserialize(i)
@WordItems << worditem_tmp
end
end
end
end
# DeleteWordItems返回参数结构体
class DeleteWordItemsResponse < TencentCloud::Common::AbstractModel
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :RequestId
def initialize(requestid=nil)
@RequestId = requestid
end
def deserialize(params)
@RequestId = params['RequestId']
end
end
# DependencyParsing请求参数结构体
class DependencyParsingRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分析的文本(仅支持UTF-8格式,不超过200字)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# DependencyParsing返回参数结构体
class DependencyParsingResponse < TencentCloud::Common::AbstractModel
# @param DpTokens: 句法依存分析结果,其中句法依存关系的类型包括:
#
主谓关系,eg: 我送她一束花 (我 <-- 送)
# 动宾关系,eg: 我送她一束花 (送 --> 花)
# 间宾关系,eg: 我送她一束花 (送 --> 她)
# 前置宾语,eg: 他什么书都读 (书 <-- 读)
# 兼语,eg: 他请我吃饭 (请 --> 我)
# 定中关系,eg: 红苹果 (红 <-- 苹果)
# 状中结构,eg: 非常美丽 (非常 <-- 美丽)
# 动补结构,eg: 做完了作业 (做 --> 完)
# 并列关系,eg: 大山和大海 (大山 --> 大海)
# 介宾关系,eg: 在贸易区内 (在 --> 内)
# 左附加关系,eg: 大山和大海 (和 <-- 大海)
# 右附加关系,eg: 孩子们 (孩子 --> 们)
# 独立结构,eg: 两个单句在结构上彼此独立
# 标点符号,eg: 。
# 核心关系,eg: 整个句子的核心
# @type DpTokens: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :DpTokens, :RequestId
def initialize(dptokens=nil, requestid=nil)
@DpTokens = dptokens
@RequestId = requestid
end
def deserialize(params)
unless params['DpTokens'].nil?
@DpTokens = []
params['DpTokens'].each do |i|
dptoken_tmp = DpToken.new
dptoken_tmp.deserialize(i)
@DpTokens << dptoken_tmp
end
end
@RequestId = params['RequestId']
end
end
# DescribeDict请求参数结构体
class DescribeDictRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param Name: 自定义词库名称,模糊搜索。
# @type Name: String
attr_accessor :DictId, :Name
def initialize(dictid=nil, name=nil)
@DictId = dictid
@Name = name
end
def deserialize(params)
@DictId = params['DictId']
@Name = params['Name']
end
end
# DescribeDict返回参数结构体
class DescribeDictResponse < TencentCloud::Common::AbstractModel
# @param Dicts: 查询到的词库信息列表。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Dicts: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Dicts, :RequestId
def initialize(dicts=nil, requestid=nil)
@Dicts = dicts
@RequestId = requestid
end
def deserialize(params)
unless params['Dicts'].nil?
@Dicts = []
params['Dicts'].each do |i|
dictinfo_tmp = DictInfo.new
dictinfo_tmp.deserialize(i)
@Dicts << dictinfo_tmp
end
end
@RequestId = params['RequestId']
end
end
# DescribeDicts请求参数结构体
class DescribeDictsRequest < TencentCloud::Common::AbstractModel
# @param Limit: 每页数据量,范围为1~100,默认为10。
# @type Limit: Integer
# @param Offset: 分页偏移量,从0开始,默认为0。
# @type Offset: Integer
attr_accessor :Limit, :Offset
def initialize(limit=nil, offset=nil)
@Limit = limit
@Offset = offset
end
def deserialize(params)
@Limit = params['Limit']
@Offset = params['Offset']
end
end
# DescribeDicts返回参数结构体
class DescribeDictsResponse < TencentCloud::Common::AbstractModel
# @param TotalCount: 记录总条数。
# @type TotalCount: Integer
# @param Dicts: 自定义词库信息列表。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Dicts: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :TotalCount, :Dicts, :RequestId
def initialize(totalcount=nil, dicts=nil, requestid=nil)
@TotalCount = totalcount
@Dicts = dicts
@RequestId = requestid
end
def deserialize(params)
@TotalCount = params['TotalCount']
unless params['Dicts'].nil?
@Dicts = []
params['Dicts'].each do |i|
dictinfo_tmp = DictInfo.new
dictinfo_tmp.deserialize(i)
@Dicts << dictinfo_tmp
end
end
@RequestId = params['RequestId']
end
end
# DescribeWordItems请求参数结构体
class DescribeWordItemsRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param Offset: 分页偏移量,从0开始,默认为0。
# @type Offset: Integer
# @param Limit: 每页数据量,范围为1~100,默认为10。
# @type Limit: Integer
# @param Text: 待检索的词条文本,支持模糊匹配。
# @type Text: String
attr_accessor :DictId, :Offset, :Limit, :Text
def initialize(dictid=nil, offset=nil, limit=nil, text=nil)
@DictId = dictid
@Offset = offset
@Limit = limit
@Text = text
end
def deserialize(params)
@DictId = params['DictId']
@Offset = params['Offset']
@Limit = params['Limit']
@Text = params['Text']
end
end
# DescribeWordItems返回参数结构体
class DescribeWordItemsResponse < TencentCloud::Common::AbstractModel
# @param TotalCount: 词条记录总条数。
# @type TotalCount: Integer
# @param WordItems: 词条信息列表。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type WordItems: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :TotalCount, :WordItems, :RequestId
def initialize(totalcount=nil, worditems=nil, requestid=nil)
@TotalCount = totalcount
@WordItems = worditems
@RequestId = requestid
end
def deserialize(params)
@TotalCount = params['TotalCount']
unless params['WordItems'].nil?
@WordItems = []
params['WordItems'].each do |i|
worditem_tmp = WordItem.new
worditem_tmp.deserialize(i)
@WordItems << worditem_tmp
end
end
@RequestId = params['RequestId']
end
end
# 自定义词库信息
class DictInfo < TencentCloud::Common::AbstractModel
# @param Name: 自定义词库名称。
# @type Name: String
# @param Id: 自定义词库ID。
# @type Id: String
# @param Description: 自定义词库描述信息。
# @type Description: String
# @param UpdateTime: 自定义词库修改时间,形式为:yyyy-mm-dd hh:mm:ss。
# @type UpdateTime: String
# @param CreateTime: 自定义词库创建时间,形式为:yyyy-mm-dd hh:mm:ss。
# @type CreateTime: String
attr_accessor :Name, :Id, :Description, :UpdateTime, :CreateTime
def initialize(name=nil, id=nil, description=nil, updatetime=nil, createtime=nil)
@Name = name
@Id = id
@Description = description
@UpdateTime = updatetime
@CreateTime = createtime
end
def deserialize(params)
@Name = params['Name']
@Id = params['Id']
@Description = params['Description']
@UpdateTime = params['UpdateTime']
@CreateTime = params['CreateTime']
end
end
# 句法依存分析结果,包括基础词,基础词的序号,当前词父节点的序号,句法依存关系的类型
class DpToken < TencentCloud::Common::AbstractModel
# @param Relation: 句法依存关系的类型
# @type Relation: String
# @param HeadId: 当前词父节点的序号
# @type HeadId: Integer
# @param Word: 基础词
# @type Word: String
# @param Id: 基础词的序号
# @type Id: Integer
attr_accessor :Relation, :HeadId, :Word, :Id
def initialize(relation=nil, headid=nil, word=nil, id=nil)
@Relation = relation
@HeadId = headid
@Word = word
@Id = id
end
def deserialize(params)
@Relation = params['Relation']
@HeadId = params['HeadId']
@Word = params['Word']
@Id = params['Id']
end
end
# 文本润色结果
class Embellish < TencentCloud::Common::AbstractModel
# @param Text: 润色后的文本。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Text: String
# @param EmbellishType: 润色类型。类型如下:
# expansion:扩写
# rewriting:改写
# translation_m2a:从现代文改写为古文
# translation_a2m:从古文改写为现代文
# 注意:此字段可能返回 null,表示取不到有效值。
# @type EmbellishType: String
attr_accessor :Text, :EmbellishType
def initialize(text=nil, embellishtype=nil)
@Text = text
@EmbellishType = embellishtype
end
def deserialize(params)
@Text = params['Text']
@EmbellishType = params['EmbellishType']
end
end
# 实体识别结果。
class Entity < TencentCloud::Common::AbstractModel
# @param Word: 基础词。
# @type Word: String
# @param BeginOffset: 基础词在NormalText中的起始位置。
# @type BeginOffset: Integer
# @param Length: 基础词的长度。
# @type Length: Integer
# @param Type: 实体类型的标准名字。
# @type Type: String
# @param Name: 类型名字的自然语言表达。(中文或英文)
# @type Name: String
attr_accessor :Word, :BeginOffset, :Length, :Type, :Name
def initialize(word=nil, beginoffset=nil, length=nil, type=nil, name=nil)
@Word = word
@BeginOffset = beginoffset
@Length = length
@Type = type
@Name = name
end
def deserialize(params)
@Word = params['Word']
@BeginOffset = params['BeginOffset']
@Length = params['Length']
@Type = params['Type']
@Name = params['Name']
end
end
# EvaluateSentenceSimilarity请求参数结构体
class EvaluateSentenceSimilarityRequest < TencentCloud::Common::AbstractModel
# @param SentencePairList: 待分析的句子对数组。句子对应不超过1对,仅支持中文文本,原句子与目标句子均应不超过64字符。
# @type SentencePairList: Array
attr_accessor :SentencePairList
def initialize(sentencepairlist=nil)
@SentencePairList = sentencepairlist
end
def deserialize(params)
unless params['SentencePairList'].nil?
@SentencePairList = []
params['SentencePairList'].each do |i|
sentencepair_tmp = SentencePair.new
sentencepair_tmp.deserialize(i)
@SentencePairList << sentencepair_tmp
end
end
end
end
# EvaluateSentenceSimilarity返回参数结构体
class EvaluateSentenceSimilarityResponse < TencentCloud::Common::AbstractModel
# @param ScoreList: 每个句子对的相似度分值。
# @type ScoreList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :ScoreList, :RequestId
def initialize(scorelist=nil, requestid=nil)
@ScoreList = scorelist
@RequestId = requestid
end
def deserialize(params)
@ScoreList = params['ScoreList']
@RequestId = params['RequestId']
end
end
# EvaluateWordSimilarity请求参数结构体
class EvaluateWordSimilarityRequest < TencentCloud::Common::AbstractModel
# @param SourceWord: 计算相似度的源词。(仅支持UTF-8格式,不超过10字符)
# @type SourceWord: String
# @param TargetWord: 计算相似度的目标词。(仅支持UTF-8格式,不超过10字符)
# @type TargetWord: String
attr_accessor :SourceWord, :TargetWord
def initialize(sourceword=nil, targetword=nil)
@SourceWord = sourceword
@TargetWord = targetword
end
def deserialize(params)
@SourceWord = params['SourceWord']
@TargetWord = params['TargetWord']
end
end
# EvaluateWordSimilarity返回参数结构体
class EvaluateWordSimilarityResponse < TencentCloud::Common::AbstractModel
# @param Similarity: 词相似度分值。
# @type Similarity: Float
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Similarity, :RequestId
def initialize(similarity=nil, requestid=nil)
@Similarity = similarity
@RequestId = requestid
end
def deserialize(params)
@Similarity = params['Similarity']
@RequestId = params['RequestId']
end
end
# GenerateCouplet请求参数结构体
class GenerateCoupletRequest < TencentCloud::Common::AbstractModel
# @param Text: 生成对联的关键词。长度需>=2,当长度>2时,自动截取前两个字作为关键字。内容需为常用汉字(不含有数字、英文、韩语、日语、符号等等其他)。
# @type Text: String
# @param TargetType: 返回的文本结果为繁体还是简体。0:简体;1:繁体。默认为0。
# @type TargetType: Integer
attr_accessor :Text, :TargetType
def initialize(text=nil, targettype=nil)
@Text = text
@TargetType = targettype
end
def deserialize(params)
@Text = params['Text']
@TargetType = params['TargetType']
end
end
# GenerateCouplet返回参数结构体
class GenerateCoupletResponse < TencentCloud::Common::AbstractModel
# @param TopScroll: 横批。
# @type TopScroll: String
# @param Content: 上联与下联。
# @type Content: Array
# @param RandomCause: 当对联随机生成时,展示随机生成原因。
# @type RandomCause: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :TopScroll, :Content, :RandomCause, :RequestId
def initialize(topscroll=nil, content=nil, randomcause=nil, requestid=nil)
@TopScroll = topscroll
@Content = content
@RandomCause = randomcause
@RequestId = requestid
end
def deserialize(params)
@TopScroll = params['TopScroll']
@Content = params['Content']
@RandomCause = params['RandomCause']
@RequestId = params['RequestId']
end
end
# GenerateKeywordSentence请求参数结构体
class GenerateKeywordSentenceRequest < TencentCloud::Common::AbstractModel
# @param WordList: 生成句子的关键词,关键词个数需不超过4个,中文关键词长度应不超过10字符,英文关键词长度不超过3个单词。关键词中不可包含标点符号。
# @type WordList: Array
# @param Number: 返回生成句子的个数。数量需>=1且<=5。
# (注意实际结果可能小于指定个数)
# @type Number: Integer
# @param Domain: 指定生成句子的领域,支持领域如下:
# general:通用领域,支持中英文
# academic:学术领域,仅支持英文
# 默认为general(通用领域)。
# @type Domain: String
attr_accessor :WordList, :Number, :Domain
def initialize(wordlist=nil, number=nil, domain=nil)
@WordList = wordlist
@Number = number
@Domain = domain
end
def deserialize(params)
@WordList = params['WordList']
@Number = params['Number']
@Domain = params['Domain']
end
end
# GenerateKeywordSentence返回参数结构体
class GenerateKeywordSentenceResponse < TencentCloud::Common::AbstractModel
# @param KeywordSentenceList: 生成的句子列表。
# @type KeywordSentenceList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :KeywordSentenceList, :RequestId
def initialize(keywordsentencelist=nil, requestid=nil)
@KeywordSentenceList = keywordsentencelist
@RequestId = requestid
end
def deserialize(params)
unless params['KeywordSentenceList'].nil?
@KeywordSentenceList = []
params['KeywordSentenceList'].each do |i|
keywordsentence_tmp = KeywordSentence.new
keywordsentence_tmp.deserialize(i)
@KeywordSentenceList << keywordsentence_tmp
end
end
@RequestId = params['RequestId']
end
end
# GeneratePoetry请求参数结构体
class GeneratePoetryRequest < TencentCloud::Common::AbstractModel
# @param Text: 生成诗词的关键词。
# @type Text: String
# @param PoetryType: 生成诗词的类型。0:藏头或藏身;1:藏头;2:藏身。默认为0。
# @type PoetryType: Integer
# @param Genre: 诗的体裁。0:五言律诗或七言律诗;5:五言律诗;7:七言律诗。默认为0。
# @type Genre: Integer
attr_accessor :Text, :PoetryType, :Genre
def initialize(text=nil, poetrytype=nil, genre=nil)
@Text = text
@PoetryType = poetrytype
@Genre = genre
end
def deserialize(params)
@Text = params['Text']
@PoetryType = params['PoetryType']
@Genre = params['Genre']
end
end
# GeneratePoetry返回参数结构体
class GeneratePoetryResponse < TencentCloud::Common::AbstractModel
# @param Title: 诗题,即输入的生成诗词的关键词。
# @type Title: String
# @param Content: 诗的内容。
# @type Content: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Title, :Content, :RequestId
def initialize(title=nil, content=nil, requestid=nil)
@Title = title
@Content = content
@RequestId = requestid
end
def deserialize(params)
@Title = params['Title']
@Content = params['Content']
@RequestId = params['RequestId']
end
end
# 关键词提取结果
class Keyword < TencentCloud::Common::AbstractModel
# @param Score: 权重
# @type Score: Float
# @param Word: 关键词
# @type Word: String
attr_accessor :Score, :Word
def initialize(score=nil, word=nil)
@Score = score
@Word = word
end
def deserialize(params)
@Score = params['Score']
@Word = params['Word']
end
end
# 通过关键词生成的句子信息
class KeywordSentence < TencentCloud::Common::AbstractModel
# @param TargetText: 通过关键词生成的句子。
# @type TargetText: String
attr_accessor :TargetText
def initialize(targettext=nil)
@TargetText = targettext
end
def deserialize(params)
@TargetText = params['TargetText']
end
end
# KeywordsExtraction请求参数结构体
class KeywordsExtractionRequest < TencentCloud::Common::AbstractModel
# @param Text: 待处理的文本(仅支持UTF-8格式,不超过10000字符)
# @type Text: String
# @param Num: 指定关键词个数上限(默认值为5)
# @type Num: Integer
attr_accessor :Text, :Num
def initialize(text=nil, num=nil)
@Text = text
@Num = num
end
def deserialize(params)
@Text = params['Text']
@Num = params['Num']
end
end
# KeywordsExtraction返回参数结构体
class KeywordsExtractionResponse < TencentCloud::Common::AbstractModel
# @param Keywords: 关键词提取结果
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Keywords: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Keywords, :RequestId
def initialize(keywords=nil, requestid=nil)
@Keywords = keywords
@RequestId = requestid
end
def deserialize(params)
unless params['Keywords'].nil?
@Keywords = []
params['Keywords'].each do |i|
keyword_tmp = Keyword.new
keyword_tmp.deserialize(i)
@Keywords << keyword_tmp
end
end
@RequestId = params['RequestId']
end
end
# LexicalAnalysis请求参数结构体
class LexicalAnalysisRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分析的文本(仅支持UTF-8格式,不超过500字)
# @type Text: String
# @param DictId: 指定要加载的自定义词库ID。
# @type DictId: String
# @param Flag: 词法分析模式(默认取2值):
# 1、高精度(混合粒度分词能力);
# 2、高性能(单粒度分词能力);
# @type Flag: Integer
attr_accessor :Text, :DictId, :Flag
def initialize(text=nil, dictid=nil, flag=nil)
@Text = text
@DictId = dictid
@Flag = flag
end
def deserialize(params)
@Text = params['Text']
@DictId = params['DictId']
@Flag = params['Flag']
end
end
# LexicalAnalysis返回参数结构体
class LexicalAnalysisResponse < TencentCloud::Common::AbstractModel
# @param NerTokens: 命名实体识别结果。取值范围:
# PER:表示人名,如刘德华、贝克汉姆
# LOC:表示地名,如北京、华山
# ORG:表示机构团体名,如腾讯、最高人民法院、人大附中
# PRODUCTION:表示产品名,如QQ、微信、iPhone
# 注意:此字段可能返回 null,表示取不到有效值。
# @type NerTokens: Array
# @param PosTokens: 分词&词性标注结果(词性表请参见附录)
# @type PosTokens: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :NerTokens, :PosTokens, :RequestId
def initialize(nertokens=nil, postokens=nil, requestid=nil)
@NerTokens = nertokens
@PosTokens = postokens
@RequestId = requestid
end
def deserialize(params)
unless params['NerTokens'].nil?
@NerTokens = []
params['NerTokens'].each do |i|
nertoken_tmp = NerToken.new
nertoken_tmp.deserialize(i)
@NerTokens << nertoken_tmp
end
end
unless params['PosTokens'].nil?
@PosTokens = []
params['PosTokens'].each do |i|
postoken_tmp = PosToken.new
postoken_tmp.deserialize(i)
@PosTokens << postoken_tmp
end
end
@RequestId = params['RequestId']
end
end
# 命名实体识别结果
class NerToken < TencentCloud::Common::AbstractModel
# @param Word: 基础词
# @type Word: String
# @param Length: 长度
# @type Length: Integer
# @param BeginOffset: 起始位置
# @type BeginOffset: Integer
# @param Type: 命名实体类型
# @type Type: String
attr_accessor :Word, :Length, :BeginOffset, :Type
def initialize(word=nil, length=nil, beginoffset=nil, type=nil)
@Word = word
@Length = length
@BeginOffset = beginoffset
@Type = type
end
def deserialize(params)
@Word = params['Word']
@Length = params['Length']
@BeginOffset = params['BeginOffset']
@Type = params['Type']
end
end
# ParseWords请求参数结构体
class ParseWordsRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分析的文本(支持中英文文本,不超过500字符)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# ParseWords返回参数结构体
class ParseWordsResponse < TencentCloud::Common::AbstractModel
# @param NormalText: 输入文本正则化的结果。(包括对英文文本中的开头和实体进行大写等)
# @type NormalText: String
# @param BasicParticiples: 基础粒度分词和词性标注的结果。(请参见附录[词性表](https://cloud.tencent.com/document/product/271/36460))
# @type BasicParticiples: Array
# @param CompoundParticiples: 复合粒度分词和词性标注的结果。(请参见附录[词性表](https://cloud.tencent.com/document/product/271/36460))
# @type CompoundParticiples: Array
# @param Entities: 实体识别结果。(请参见附录[实体类型数据](https://cloud.tencent.com/document/product/271/90592))
# @type Entities: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :NormalText, :BasicParticiples, :CompoundParticiples, :Entities, :RequestId
def initialize(normaltext=nil, basicparticiples=nil, compoundparticiples=nil, entities=nil, requestid=nil)
@NormalText = normaltext
@BasicParticiples = basicparticiples
@CompoundParticiples = compoundparticiples
@Entities = entities
@RequestId = requestid
end
def deserialize(params)
@NormalText = params['NormalText']
unless params['BasicParticiples'].nil?
@BasicParticiples = []
params['BasicParticiples'].each do |i|
basicparticiple_tmp = BasicParticiple.new
basicparticiple_tmp.deserialize(i)
@BasicParticiples << basicparticiple_tmp
end
end
unless params['CompoundParticiples'].nil?
@CompoundParticiples = []
params['CompoundParticiples'].each do |i|
compoundparticiple_tmp = CompoundParticiple.new
compoundparticiple_tmp.deserialize(i)
@CompoundParticiples << compoundparticiple_tmp
end
end
unless params['Entities'].nil?
@Entities = []
params['Entities'].each do |i|
entity_tmp = Entity.new
entity_tmp.deserialize(i)
@Entities << entity_tmp
end
end
@RequestId = params['RequestId']
end
end
# 分词&词性标注结果
class PosToken < TencentCloud::Common::AbstractModel
# @param Word: 基础词
# @type Word: String
# @param Length: 长度
# @type Length: Integer
# @param BeginOffset: 起始位置
# @type BeginOffset: Integer
# @param Pos: 词性
# @type Pos: String
attr_accessor :Word, :Length, :BeginOffset, :Pos
def initialize(word=nil, length=nil, beginoffset=nil, pos=nil)
@Word = word
@Length = length
@BeginOffset = beginoffset
@Pos = pos
end
def deserialize(params)
@Word = params['Word']
@Length = params['Length']
@BeginOffset = params['BeginOffset']
@Pos = params['Pos']
end
end
# RetrieveSimilarWords请求参数结构体
class RetrieveSimilarWordsRequest < TencentCloud::Common::AbstractModel
# @param Text: 输入的词语。(仅支持UTF-8格式,不超过10字符)
# @type Text: String
# @param Number: 召回的相似词个数,取值范围为1-20。
# @type Number: Integer
attr_accessor :Text, :Number
def initialize(text=nil, number=nil)
@Text = text
@Number = number
end
def deserialize(params)
@Text = params['Text']
@Number = params['Number']
end
end
# RetrieveSimilarWords返回参数结构体
class RetrieveSimilarWordsResponse < TencentCloud::Common::AbstractModel
# @param WordList: 召回的相似词数组。
# @type WordList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :WordList, :RequestId
def initialize(wordlist=nil, requestid=nil)
@WordList = wordlist
@RequestId = requestid
end
def deserialize(params)
@WordList = params['WordList']
@RequestId = params['RequestId']
end
end
# 词条搜索的结果,主要描述该词条是否存在以及相关的词性。
class SearchResult < TencentCloud::Common::AbstractModel
# @param Text: 被搜索的词条文本。
# @type Text: String
# @param IsExist: 0表示词条不存在,1表示存在。
# @type IsExist: Integer
# @param MatchText: 匹配到的词条文本。
# @type MatchText: String
# @param Pos: 词条的词性。
# @type Pos: String
attr_accessor :Text, :IsExist, :MatchText, :Pos
def initialize(text=nil, isexist=nil, matchtext=nil, pos=nil)
@Text = text
@IsExist = isexist
@MatchText = matchtext
@Pos = pos
end
def deserialize(params)
@Text = params['Text']
@IsExist = params['IsExist']
@MatchText = params['MatchText']
@Pos = params['Pos']
end
end
# SearchWordItems请求参数结构体
class SearchWordItemsRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param WordItems: 待检索的词条集合。
# @type WordItems: Array
attr_accessor :DictId, :WordItems
def initialize(dictid=nil, worditems=nil)
@DictId = dictid
@WordItems = worditems
end
def deserialize(params)
@DictId = params['DictId']
unless params['WordItems'].nil?
@WordItems = []
params['WordItems'].each do |i|
worditem_tmp = WordItem.new
worditem_tmp.deserialize(i)
@WordItems << worditem_tmp
end
end
end
end
# SearchWordItems返回参数结构体
class SearchWordItemsResponse < TencentCloud::Common::AbstractModel
# @param Results: 词条检索结果集合。
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Results: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Results, :RequestId
def initialize(results=nil, requestid=nil)
@Results = results
@RequestId = requestid
end
def deserialize(params)
unless params['Results'].nil?
@Results = []
params['Results'].each do |i|
searchresult_tmp = SearchResult.new
searchresult_tmp.deserialize(i)
@Results << searchresult_tmp
end
end
@RequestId = params['RequestId']
end
end
# SentenceCorrection请求参数结构体
class SentenceCorrectionRequest < TencentCloud::Common::AbstractModel
# @param TextList: 待纠错的句子列表。可以以数组方式在一次请求中填写多个待纠错的句子。文本统一使用utf-8格式编码,每个中文句子的长度不超过150字符,每个英文句子的长度不超过100个单词,且数组长度需小于30,即句子总数需少于30句。
# @type TextList: Array
attr_accessor :TextList
def initialize(textlist=nil)
@TextList = textlist
end
def deserialize(params)
@TextList = params['TextList']
end
end
# SentenceCorrection返回参数结构体
class SentenceCorrectionResponse < TencentCloud::Common::AbstractModel
# @param CorrectionList: 纠错结果列表。
# (注意仅展示错误句子的纠错结果,若句子无错则不展示,若全部待纠错句子都被认为无错,则可能返回数组为空)
# @type CorrectionList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :CorrectionList, :RequestId
def initialize(correctionlist=nil, requestid=nil)
@CorrectionList = correctionlist
@RequestId = requestid
end
def deserialize(params)
unless params['CorrectionList'].nil?
@CorrectionList = []
params['CorrectionList'].each do |i|
correctionitem_tmp = CorrectionItem.new
correctionitem_tmp.deserialize(i)
@CorrectionList << correctionitem_tmp
end
end
@RequestId = params['RequestId']
end
end
# SentenceEmbedding请求参数结构体
class SentenceEmbeddingRequest < TencentCloud::Common::AbstractModel
# @param Text: 输入的文本(仅支持UTF-8格式,不超过500字)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# SentenceEmbedding返回参数结构体
class SentenceEmbeddingResponse < TencentCloud::Common::AbstractModel
# @param Vector: 句向量数组
# @type Vector: Array
# @param Dimension: 句向量的维度
# @type Dimension: Integer
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Vector, :Dimension, :RequestId
def initialize(vector=nil, dimension=nil, requestid=nil)
@Vector = vector
@Dimension = dimension
@RequestId = requestid
end
def deserialize(params)
@Vector = params['Vector']
@Dimension = params['Dimension']
@RequestId = params['RequestId']
end
end
# 待分析的句子对
class SentencePair < TencentCloud::Common::AbstractModel
# @param SourceText: 需要与目标句子计算相似度的源句子。(仅支持UTF-8格式,不超过500字符)
# @type SourceText: String
# @param TargetText: 目标句子。(仅支持UTF-8格式,不超过500字符)
# @type TargetText: String
attr_accessor :SourceText, :TargetText
def initialize(sourcetext=nil, targettext=nil)
@SourceText = sourcetext
@TargetText = targettext
end
def deserialize(params)
@SourceText = params['SourceText']
@TargetText = params['TargetText']
end
end
# SentimentAnalysis请求参数结构体
class SentimentAnalysisRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分析的文本(仅支持UTF-8格式,不超过200字)
# @type Text: String
# @param Flag: 待分析文本所属的类型,仅当输入参数Mode取值为2class时有效(默认取4值):
# 1、商品评论类
# 2、社交类
# 3、美食酒店类
# 4、通用领域类
# @type Flag: Integer
# @param Mode: 情感分类模式选项,可取2class或3class(默认值为2class)
# 1、2class:返回正负面二分类情感结果
# 2、3class:返回正负面及中性三分类情感结果
# @type Mode: String
attr_accessor :Text, :Flag, :Mode
def initialize(text=nil, flag=nil, mode=nil)
@Text = text
@Flag = flag
@Mode = mode
end
def deserialize(params)
@Text = params['Text']
@Flag = params['Flag']
@Mode = params['Mode']
end
end
# SentimentAnalysis返回参数结构体
class SentimentAnalysisResponse < TencentCloud::Common::AbstractModel
# @param Positive: 正面情感概率
# @type Positive: Float
# @param Neutral: 中性情感概率,当输入参数Mode取值为3class时有效,否则值为空
# 注意:此字段可能返回 null,表示取不到有效值。
# @type Neutral: Float
# @param Negative: 负面情感概率
# @type Negative: Float
# @param Sentiment: 情感分类结果:
# 1、positive,表示正面情感
# 2、negative,表示负面情感
# 3、neutral,表示中性、无情感
# @type Sentiment: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Positive, :Neutral, :Negative, :Sentiment, :RequestId
def initialize(positive=nil, neutral=nil, negative=nil, sentiment=nil, requestid=nil)
@Positive = positive
@Neutral = neutral
@Negative = negative
@Sentiment = sentiment
@RequestId = requestid
end
def deserialize(params)
@Positive = params['Positive']
@Neutral = params['Neutral']
@Negative = params['Negative']
@Sentiment = params['Sentiment']
@RequestId = params['RequestId']
end
end
# SimilarWords请求参数结构体
class SimilarWordsRequest < TencentCloud::Common::AbstractModel
# @param Text: 输入的词语(仅支持UTF-8格式,不超过20字)
# @type Text: String
# @param WordNumber: 相似词个数;取值范围:1-200,默认为10;
# @type WordNumber: Integer
attr_accessor :Text, :WordNumber
def initialize(text=nil, wordnumber=nil)
@Text = text
@WordNumber = wordnumber
end
def deserialize(params)
@Text = params['Text']
@WordNumber = params['WordNumber']
end
end
# SimilarWords返回参数结构体
class SimilarWordsResponse < TencentCloud::Common::AbstractModel
# @param SimilarWords: 相似词数组
# @type SimilarWords: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :SimilarWords, :RequestId
def initialize(similarwords=nil, requestid=nil)
@SimilarWords = similarwords
@RequestId = requestid
end
def deserialize(params)
@SimilarWords = params['SimilarWords']
@RequestId = params['RequestId']
end
end
# 文本相似度
class Similarity < TencentCloud::Common::AbstractModel
# @param Text: 目标文本句子
# @type Text: String
# @param Score: 相似度分数
# @type Score: Float
attr_accessor :Text, :Score
def initialize(text=nil, score=nil)
@Text = text
@Score = score
end
def deserialize(params)
@Text = params['Text']
@Score = params['Score']
end
end
# TextClassification请求参数结构体
class TextClassificationRequest < TencentCloud::Common::AbstractModel
# @param Text: 待分类的文本(仅支持UTF-8格式,不超过10000字)
# @type Text: String
# @param Flag: 领域分类体系(默认取1值):
# 1、通用领域,二分类
# 2、新闻领域,五分类。类别数据不一定全部返回,详情见类目映射表(注意:目前五分类已下线不可用)
# @type Flag: Integer
attr_accessor :Text, :Flag
def initialize(text=nil, flag=nil)
@Text = text
@Flag = flag
end
def deserialize(params)
@Text = params['Text']
@Flag = params['Flag']
end
end
# TextClassification返回参数结构体
class TextClassificationResponse < TencentCloud::Common::AbstractModel
# @param Classes: 文本分类结果(文本分类映射表请参见附录)
# @type Classes: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Classes, :RequestId
def initialize(classes=nil, requestid=nil)
@Classes = classes
@RequestId = requestid
end
def deserialize(params)
unless params['Classes'].nil?
@Classes = []
params['Classes'].each do |i|
classificationresult_tmp = ClassificationResult.new
classificationresult_tmp.deserialize(i)
@Classes << classificationresult_tmp
end
end
@RequestId = params['RequestId']
end
end
# TextCorrectionPro请求参数结构体
class TextCorrectionProRequest < TencentCloud::Common::AbstractModel
# @param Text: 待纠错的文本(仅支持UTF-8格式,不超过128字符)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# TextCorrectionPro返回参数结构体
class TextCorrectionProResponse < TencentCloud::Common::AbstractModel
# @param CCITokens: 纠错详情
# 注意:此字段可能返回 null,表示取不到有效值。
# @type CCITokens: Array
# @param ResultText: 纠错后的文本
# @type ResultText: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :CCITokens, :ResultText, :RequestId
def initialize(ccitokens=nil, resulttext=nil, requestid=nil)
@CCITokens = ccitokens
@ResultText = resulttext
@RequestId = requestid
end
def deserialize(params)
unless params['CCITokens'].nil?
@CCITokens = []
params['CCITokens'].each do |i|
ccitoken_tmp = CCIToken.new
ccitoken_tmp.deserialize(i)
@CCITokens << ccitoken_tmp
end
end
@ResultText = params['ResultText']
@RequestId = params['RequestId']
end
end
# TextCorrection请求参数结构体
class TextCorrectionRequest < TencentCloud::Common::AbstractModel
# @param Text: 待纠错的文本(仅支持UTF-8格式,不超过2000字符)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# TextCorrection返回参数结构体
class TextCorrectionResponse < TencentCloud::Common::AbstractModel
# @param CCITokens: 纠错详情
# 注意:此字段可能返回 null,表示取不到有效值。
# @type CCITokens: Array
# @param ResultText: 纠错后的文本
# @type ResultText: String
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :CCITokens, :ResultText, :RequestId
def initialize(ccitokens=nil, resulttext=nil, requestid=nil)
@CCITokens = ccitokens
@ResultText = resulttext
@RequestId = requestid
end
def deserialize(params)
unless params['CCITokens'].nil?
@CCITokens = []
params['CCITokens'].each do |i|
ccitoken_tmp = CCIToken.new
ccitoken_tmp.deserialize(i)
@CCITokens << ccitoken_tmp
end
end
@ResultText = params['ResultText']
@RequestId = params['RequestId']
end
end
# TextEmbellish请求参数结构体
class TextEmbellishRequest < TencentCloud::Common::AbstractModel
# @param Text: 待润色的文本。中文文本长度需<=50字符;英文文本长度需<=30个单词。
# @type Text: String
# @param SourceLang: 待润色文本的语言类型,支持语言如下:
# zh:中文
# en:英文
# @type SourceLang: String
# @param Number: 返回润色结果的个数。数量需>=1且<=5。
# (注意实际结果可能小于指定个数)
# @type Number: Integer
# @param Style: 控制润色类型,类型如下:
# both:同时返回改写和扩写
# expansion:扩写
# rewriting:改写
# m2a:从现代文改写为古文
# a2m:从古文改写为现代文
# 默认为both。
# @type Style: String
attr_accessor :Text, :SourceLang, :Number, :Style
def initialize(text=nil, sourcelang=nil, number=nil, style=nil)
@Text = text
@SourceLang = sourcelang
@Number = number
@Style = style
end
def deserialize(params)
@Text = params['Text']
@SourceLang = params['SourceLang']
@Number = params['Number']
@Style = params['Style']
end
end
# TextEmbellish返回参数结构体
class TextEmbellishResponse < TencentCloud::Common::AbstractModel
# @param EmbellishList: 润色结果列表。
# @type EmbellishList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :EmbellishList, :RequestId
def initialize(embellishlist=nil, requestid=nil)
@EmbellishList = embellishlist
@RequestId = requestid
end
def deserialize(params)
unless params['EmbellishList'].nil?
@EmbellishList = []
params['EmbellishList'].each do |i|
embellish_tmp = Embellish.new
embellish_tmp.deserialize(i)
@EmbellishList << embellish_tmp
end
end
@RequestId = params['RequestId']
end
end
# TextSimilarityPro请求参数结构体
class TextSimilarityProRequest < TencentCloud::Common::AbstractModel
# @param SrcText: 需要与目标句子计算相似度的源句子(仅支持UTF-8格式,不超过128字符)
# @type SrcText: String
# @param TargetText: 目标句子(仅支持UTF-8格式,不超过128字符)
# @type TargetText: Array
attr_accessor :SrcText, :TargetText
def initialize(srctext=nil, targettext=nil)
@SrcText = srctext
@TargetText = targettext
end
def deserialize(params)
@SrcText = params['SrcText']
@TargetText = params['TargetText']
end
end
# TextSimilarityPro返回参数结构体
class TextSimilarityProResponse < TencentCloud::Common::AbstractModel
# @param Similarity: 每个目标句子与源句子的相似度分值,按照分值降序排列
# @type Similarity: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Similarity, :RequestId
def initialize(similarity=nil, requestid=nil)
@Similarity = similarity
@RequestId = requestid
end
def deserialize(params)
unless params['Similarity'].nil?
@Similarity = []
params['Similarity'].each do |i|
similarity_tmp = Similarity.new
similarity_tmp.deserialize(i)
@Similarity << similarity_tmp
end
end
@RequestId = params['RequestId']
end
end
# TextSimilarity请求参数结构体
class TextSimilarityRequest < TencentCloud::Common::AbstractModel
# @param SrcText: 需要与目标句子计算相似度的源句子(仅支持UTF-8格式,不超过500字符)
# @type SrcText: String
# @param TargetText: 目标句子(以句子数量为单位消耗资源包)
# @type TargetText: Array
attr_accessor :SrcText, :TargetText
def initialize(srctext=nil, targettext=nil)
@SrcText = srctext
@TargetText = targettext
end
def deserialize(params)
@SrcText = params['SrcText']
@TargetText = params['TargetText']
end
end
# TextSimilarity返回参数结构体
class TextSimilarityResponse < TencentCloud::Common::AbstractModel
# @param Similarity: 每个目标句子与源句子的相似度分值,按照分值降序排列
# @type Similarity: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Similarity, :RequestId
def initialize(similarity=nil, requestid=nil)
@Similarity = similarity
@RequestId = requestid
end
def deserialize(params)
unless params['Similarity'].nil?
@Similarity = []
params['Similarity'].each do |i|
similarity_tmp = Similarity.new
similarity_tmp.deserialize(i)
@Similarity << similarity_tmp
end
end
@RequestId = params['RequestId']
end
end
# TextWriting请求参数结构体
class TextWritingRequest < TencentCloud::Common::AbstractModel
# @param Text: 待续写的句子,文本统一使用utf-8格式编码,长度不超过200字符。
# @type Text: String
# @param SourceLang: 待续写文本的语言类型,支持语言如下:
# zh:中文
# en:英文
# @type SourceLang: String
# @param Number: 返回续写结果的个数。数量需>=1且<=5。
# (注意实际结果可能小于指定个数)
# @type Number: Integer
# @param Domain: 指定续写领域,支持领域如下:
# general:通用领域,支持中英文补全
# academic:学术领域,仅支持英文补全
# 默认为general(通用领域)。
# @type Domain: String
# @param Style: 指定续写风格,支持风格如下:
# science_fiction:科幻
# military_history:军事
# xuanhuan_wuxia:武侠
# urban_officialdom:职场
# 默认为xuanhuan_wuxia(武侠)。
# @type Style: String
attr_accessor :Text, :SourceLang, :Number, :Domain, :Style
def initialize(text=nil, sourcelang=nil, number=nil, domain=nil, style=nil)
@Text = text
@SourceLang = sourcelang
@Number = number
@Domain = domain
@Style = style
end
def deserialize(params)
@Text = params['Text']
@SourceLang = params['SourceLang']
@Number = params['Number']
@Domain = params['Domain']
@Style = params['Style']
end
end
# TextWriting返回参数结构体
class TextWritingResponse < TencentCloud::Common::AbstractModel
# @param WritingList: 续写结果列表。
# @type WritingList: Array
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :WritingList, :RequestId
def initialize(writinglist=nil, requestid=nil)
@WritingList = writinglist
@RequestId = requestid
end
def deserialize(params)
unless params['WritingList'].nil?
@WritingList = []
params['WritingList'].each do |i|
writing_tmp = Writing.new
writing_tmp.deserialize(i)
@WritingList << writing_tmp
end
end
@RequestId = params['RequestId']
end
end
# UpdateDict请求参数结构体
class UpdateDictRequest < TencentCloud::Common::AbstractModel
# @param DictId: 自定义词库ID。
# @type DictId: String
# @param Description: 词库描述,不超过100字。
# @type Description: String
# @param Name: 词库名称,不超过20字。
# @type Name: String
attr_accessor :DictId, :Description, :Name
def initialize(dictid=nil, description=nil, name=nil)
@DictId = dictid
@Description = description
@Name = name
end
def deserialize(params)
@DictId = params['DictId']
@Description = params['Description']
@Name = params['Name']
end
end
# UpdateDict返回参数结构体
class UpdateDictResponse < TencentCloud::Common::AbstractModel
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :RequestId
def initialize(requestid=nil)
@RequestId = requestid
end
def deserialize(params)
@RequestId = params['RequestId']
end
end
# WordEmbedding请求参数结构体
class WordEmbeddingRequest < TencentCloud::Common::AbstractModel
# @param Text: 输入的词语(仅支持UTF-8格式,不超过20字)
# @type Text: String
attr_accessor :Text
def initialize(text=nil)
@Text = text
end
def deserialize(params)
@Text = params['Text']
end
end
# WordEmbedding返回参数结构体
class WordEmbeddingResponse < TencentCloud::Common::AbstractModel
# @param Vector: 词向量数组
# @type Vector: Array
# @param Dimension: 词向量的维度
# @type Dimension: Integer
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Vector, :Dimension, :RequestId
def initialize(vector=nil, dimension=nil, requestid=nil)
@Vector = vector
@Dimension = dimension
@RequestId = requestid
end
def deserialize(params)
@Vector = params['Vector']
@Dimension = params['Dimension']
@RequestId = params['RequestId']
end
end
# 词条信息。
class WordItem < TencentCloud::Common::AbstractModel
# @param Text: 词条文本内容。
# @type Text: String
# @param CreateTime: 词条创建时间。
# @type CreateTime: String
# @param Pos: 词条的词性。
# @type Pos: String
attr_accessor :Text, :CreateTime, :Pos
def initialize(text=nil, createtime=nil, pos=nil)
@Text = text
@CreateTime = createtime
@Pos = pos
end
def deserialize(params)
@Text = params['Text']
@CreateTime = params['CreateTime']
@Pos = params['Pos']
end
end
# WordSimilarity请求参数结构体
class WordSimilarityRequest < TencentCloud::Common::AbstractModel
# @param SrcWord: 计算相似度的源词(仅支持UTF-8格式,不超过20字)
# @type SrcWord: String
# @param TargetWord: 计算相似度的目标词(仅支持UTF-8格式,不超过20字)
# @type TargetWord: String
attr_accessor :SrcWord, :TargetWord
def initialize(srcword=nil, targetword=nil)
@SrcWord = srcword
@TargetWord = targetword
end
def deserialize(params)
@SrcWord = params['SrcWord']
@TargetWord = params['TargetWord']
end
end
# WordSimilarity返回参数结构体
class WordSimilarityResponse < TencentCloud::Common::AbstractModel
# @param Similarity: 两个词语的相似度
# @type Similarity: Float
# @param RequestId: 唯一请求 ID,每次请求都会返回。定位问题时需要提供该次请求的 RequestId。
# @type RequestId: String
attr_accessor :Similarity, :RequestId
def initialize(similarity=nil, requestid=nil)
@Similarity = similarity
@RequestId = requestid
end
def deserialize(params)
@Similarity = params['Similarity']
@RequestId = params['RequestId']
end
end
# 文本续写结果
class Writing < TencentCloud::Common::AbstractModel
# @param TargetText: 续写的文本。
# @type TargetText: String
# @param PrefixText: 续写的前缀。
# @type PrefixText: String
attr_accessor :TargetText, :PrefixText
def initialize(targettext=nil, prefixtext=nil)
@TargetText = targettext
@PrefixText = prefixtext
end
def deserialize(params)
@TargetText = params['TargetText']
@PrefixText = params['PrefixText']
end
end
end
end
end